Xavier D'Haultfœuille, Yagan Hazard, Philipp Ketz
Tue 17 Nov
econ.EM
pdf html
- interest Causal inference and identification in econometrics 0.94
- favorites near “When do common time series estimands have nonparametric causal meaning?” 0.91
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.88
Abstract
Under the exclusion restriction, independence, and monotonicity assumptions, Imbens and Angrist (1994) show that the Wald estimand is equal to the local average treatment effect (LATE). However, these assumptions entail additional restrictions, which we exploit for inference on the LATE. Specifically, we propose a conditional likelihood ratio test that we show controls asymptotic size uniformly over a large class of data-generating processes. The test has very low computational cost: it requires no simulation and involves solving at most four quadratic programming problems. Monte Carlo simulations show that our test can offer substantial power gains over the Wald and Anderson-Rubin tests, including when the instrument is weak.
Peter Hull, Michal Kolesár
Tue 17 Nov
econ.EM
pdf html
- reading economics, empirical, overview, causal
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.87
- favorites near “When Did Growth Begin? New Estimates of Productivity Growth in England from 1250 to 1870” 0.83
Abstract
Empirical researchers are often interested in the relative causal impact of individuals or institutions such as teachers, schools, hospitals, or neighborhoods. Regression-based estimates of such unit-specific parameters are often noisy; empirical Bayes (eB) methods can help isolate the signal from the estimation noise. This article provides an overview of eB methodology, showing how eB shrinkage can improve compound decisions---such as selecting individuals with the highest causal impacts---even when the working model motivating the shrinkage is misspecified. Not all analyses benefit from eB shrinkage, however, and we discuss which types of causal summaries are best estimated directly and how to do so robustly. We illustrate practical takeaways with a running analysis of public elementary and middle school value-added.
Nan Liu, Yanbo Liu, Yuya Sasaki, Yichong Zhang
Wed 18 Nov
econ.EMmath.STstat.TH
pdf html
- interest Causal inference and identification in econometrics 0.92
- reading functions, observational, treatment, observational data
- favorites near “When do common time series estimands have nonparametric causal meaning?” 0.89
Abstract
This paper studies the identification, estimation, and inference of long-term quantile treatment effects (LTQTEs) when short-term experimental data are supplemented with long-term observational data. Under a latent unconfoundedness framework, we establish identification of the LTQTE curve and derive the corresponding efficient influence functions. Building on these influence functions, we propose a semiparametrically efficient estimation procedure and develop a multiplier bootstrap method for uniform inference on the LTQTE curve, yielding valid simultaneous confidence bands over a continuum of quantile indices.
Yao Luo
Thu 19 Nov
econ.EM
pdf html
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.87
- favorites near “Measuring Price Selection in Microdata: It's Not There” 0.88
- interest Causal inference and identification in econometrics 0.89
Abstract
We develop a tail-geometric approach to nonparametric identification in repeated-measurement models based on large within-unit gaps. Differencing eliminates the common latent component, while conditioning on a large gap pushes the idiosyncratic errors into opposite tails. Under a regular thinner-than-Gaussian tail condition, an idiosyncratic component concentrates around a deterministic, gap-dependent location. Consequently, the corresponding observed measurement, after recentering, reveals the distribution of the common latent component up to location. The idiosyncratic distribution can then be recovered through deconvolution. This principle applies both to the classical two-measurement Kotlarski model and to settings where only order statistics are observed. Applications include auctions and wage-offer models, whether all offers are observed or only the top-ranked ones are available.
Ziyi Liu, Yiqing Xu
Tue 17 Nov
econ.EMstat.ME
pdf html
- favorites near “When do common time series estimands have nonparametric causal meaning?” 0.89
- interest Causal inference and identification in econometrics 0.90
- reading counterfactual, units, procedures, control
Abstract
We develop conformal inference for panel counterfactual estimators using the exchangeability of prediction errors across units rather than over time. Existing conformal methods for panels with few treated units rely on temporal properties of treated-unit residuals, which are vulnerable to strong temporal heterogeneity such as nonstationarity. We instead hold out each control unit, hide its relevant outcomes, and predict them using the same rule applied to the treated unit. Target-date estimators, including difference-in-differences and synthetic control, use controls' pre-treatment and target-date outcomes to construct the counterfactual, allowing both full conformal confidence sets and Jackknife+ intervals. Full-path estimators, including factor-model imputation and matrix completion, also use controls' outcomes at other post-treatment dates; for these, we use the Jackknife+. Both procedures yield finite-sample, period-wise confidence sets under this across-unit exchangeability, without requiring stationarity or random assignment. We characterize when the coverage holds and show why a common prediction rule alone does not ensure it. For multiple treated units, a block Jackknife+ covers their average effect, while a relabeling test is exact for sharp nulls over all treated cells. We show that the synthetic control placebo test is a full conformal test of no effect that extends to full-path estimators. We also relate our procedures to a parametric bootstrap and classify common estimators by the inference they allow. We illustrate our procedures in three empirical applications.
Byunghoon Kang
Mon 16 Nov
econ.EM
pdf html
- reading moment, gmm, adjustment, double
- interest Semiparametric inference and debiased machine learning 0.91
- your work near “Valid Inference with Imperfect Synthetic Data” 0.84
Abstract
We develop Neyman orthogonal scores for GMM models that remain valid when the moment conditions may fail to hold. Existing score constructions typically require the moment conditions to hold exactly, which may fail in over-identified models, where the number of moment functions exceeds the number of parameters. We construct \emph{misspecification-robust} orthogonal score that has a closed form; it is the conventional orthogonal score plus adjustment terms, for the estimation of the Jacobian and of the weighting matrix, which are proportional to the population moment and vanish under correct specification. The moment functions may depend on nonparametric or machine learning first steps, and we show that double/debiased machine learning (DML) combined with the proposed scores and cross-fitting provides uniformly valid inference for a low-dimensional GMM estimand, while the conventional DML standard errors are inconsistent when the moment conditions fail. We illustrate the results in the partially linear instrumental variables (PLIV) model and in the local average treatment effects (LATE) model with multiple instruments.
Ariel Boyarsky, Naoki Egami, Hongseok Namkoong
Fri 13 Nov
stat.MEecon.EMstat.ML
pdf html
- favorites near “SVAR (Mis)Identification and the Real Effects of Monetary Policy Shocks” 0.86
- interest Causal inference and identification in econometrics 0.91
- reading external validity, country, causal, generalize
Abstract
Assessing the external validity of experimental results is a pervasive challenge in causal inference as internally valid findings need not generalize beyond sites in which the experiments were conducted. Since the number of sites is often smaller than the number of site-level variables that moderate causal effects, we develop a partial identification framework that lets practitioners assess external validity under a wide range of assumptions that allow extrapolation across site-level variables. Our optimization-driven approach can accommodate any affine family of parameter restrictions, such as interaction terms being dominated by direct effects (the "effect hierarchy principle"). Since standard estimating-equation arguments do not apply in our setting---underspecified parameter directions responsible for partial identification make the covariance matrix singular---we develop a novel debiasing approach for estimands given by constrained optimization problems, giving rise to semiparametrically efficient estimators for our flexible partial identification bounds. We illustrate our approach on multi-site microcredit experiments harmonized by Meager (2019), whose lending environments differ widely in interest rates, loan sizes, and promotion. Holding out one country at a time, we construct external-validity intervals for the held-out country from the remaining source sites and show how our framework connects different extrapolation assumptions to concrete partial identification bounds.
Abhimanyu Gupta
Fri 13 Nov
econ.EMstat.OT
pdf html
- reading econometrics, econometric, literature, attempts
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.83
- interest Causal inference and identification in econometrics 0.88
Abstract
We conduct a non-exhaustive survey of the recent spatial econometrics literature by placing it in historical context and mapping current trends. The focus will be primarily on methodological aspects. Starting from the canonical spatial autoregressive model, we cover its many extensions and richer parameterizations. We discuss spatial panel data models and advancements in the study of nonlinear and nonparametric models. Specification testing for spatial econometric models, with its roots in the $J$-test of Davidson and MacKinnon (1981), is also covered, as is spatial correlation robust inference. Finally, we review advancements in general limit theory, limited dependent variable models, spatial weight matrix estimation and endogeneity, recent attempts at uniting the network formation and spatial literature, and sundry important topics. We take stock of the current paradigm and critically evaluate the impact of the spatial literature on applied practice.
Stephane Hess
Wed 18 Nov
econ.EM
pdf html
- reading software, consistent, units, written
- interest Causal inference and identification in econometrics 0.91
- favorites near “Advances in Nowcasting Economic Activity: Secular Trends, Large Shocks and New Data” 0.83
Abstract
Nested logit remains a widely used model, especially in applied work. However, there remains substantial confusion surrounding the existence of different versions of the model, different normalisations for the same version, and different software implementations. This paper argues that the labels in common use, RU1 and RU2, and the name non-normalised nested logit (NNNL) itself, have contributed to the confusion. The model is written in two forms: one in which utilities are divided by the structural parameter of the nest that contains them, which is consistent with utility maximisation by construction, and one without that division, known as NNNL. The difference between these forms is not one of normalisation. Normalising at the top or at the bottom gives the same model, and NNNL is itself normalised. What distinguishes the forms is whether the division is performed. In short, normalisation changes the units, whereas division changes the model. The undivided form coincides with the utility-consistent one only under specific conditions, and otherwise is consistent with utility maximisation only under utilities other than those specified, in which generic coefficients become nest-specific. The paper highlights that RU1 has been used for three different specifications, compares how widely used software packages normalise and report their parameters, and illustrates the consequences with calculated examples. The recommendation is that models be described by what they do, rather than by these labels.
Ayman Farahat
Tue 17 Nov
stat.APecon.EM
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.87
- reading researchers, large, language, treatments
- favorites near “Labor Rationing” 0.82
Abstract
Researchers increasingly regress outcomes on treatments labeled by large language models. When effects differ, OLS averages the effect on the treated (ATT) and on the untreated (ATU), weighting the smaller group more. A label's bias then has two parts: classical attenuation from misclassification, and a selective term when its errors are correlated with the outcome. In LinkedIn postings, the model misses better-paid remote jobs: a precision check passes, yet the remote pay gap is understated by nearly 40 percent, a bias that more data cannot remove. A stratified, blind audit of both labeled groups corrects it and recovers both effects.
Luke Taylor, Jesper N. Wulff
Thu 19 Nov
econ.EM
pdf html
- interest Causal inference and identification in econometrics 0.93
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.87
- favorites near “When do common time series estimands have nonparametric causal meaning?” 0.89
Abstract
Sensitivity analysis for imperfect instruments usually targets a linear two-stage least squares coefficient, yet many applications have binary outcomes and probability-scale targets. We extend the plausibly exogenous framework to the control-function probit, targeting average partial and interaction effects. We derive the generated-regressor variance, union and local-to-zero intervals, and their asymptotic coverage under stated model and calibration conditions. A violation stated on the wrong latent scale can miscalibrate the interval by a factor set by the endogeneity correlation. Applications to Catholic schooling, sibling sex composition and Confucian clans compare probit and linear analyses. Stata and R implementations accompany the analysis.
Yongzhen Feng, Xiaojun Song
Fri 13 Nov
stat.ME
pdf html
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.87
- favorites near “Difference-in-Differences with multiple time periods” 0.88
- interest Semiparametric inference and debiased machine learning 0.89
Abstract
We propose Cram\'er--von Mises- and Kolmogorov--Smirnov-type tests of constant conditional variance in nonparametric regression. The test statistics are constructed from a third-order $U$-process and remain valid when higher-order features of the conditional error distribution vary with the covariates. We derive their limiting distributions under the null hypothesis and study their asymptotic behavior under fixed alternatives and Pitman-type local alternatives. A multiplier bootstrap procedure calibrates the critical values, and Monte Carlo experiments show that the proposed tests have rejection frequencies close to the nominal level in the simulation designs. An application to the National Child Development Study provides evidence of heteroskedasticity in pooled weekly earnings. In the male hourly-earnings sample, rejecting a stronger error-distribution restriction does not entail rejecting homoskedasticity.
Louison Bocquet-Nouaille, Jérôme Morio, Benjamin Bobbia, Philippe Naveau
Fri 13 Nov
math.STstat.TH
pdf html
- reading empirical, moment, theory, control
- interest Causal inference and identification in econometrics 0.89
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.84
Abstract
Accurate estimation of Moment Generating Functions is fundamental in many applications, but empirical estimators have high variances. Within a control variate setup, Extreme Value Theory is leveraged to identify how extreme values affect the variance reduction method's optimality.
Miguel Fernandez-de-Retana, Aitor Almeida
Fri 13 Nov
stat.COcs.NAmath.NAstat.MEstat.ML
pdf html
- reading hmc, hamiltonian monte, hamiltonian, empirical
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.90
- interest Causal inference and identification in econometrics 0.88
Abstract
Hierarchical models are central to applied Bayesian statistics, yet the variance structures they depend on can produce posterior geometries that remain difficult to explore even with Hamiltonian Monte Carlo (HMC), a designation now covering a large and increasingly heterogeneous family of algorithms. This survey uses Generalized HMC as an organizing framework to describe the many variants of HMC, their methodological and computational properties, and their empirical consequences. Varying the trajectory length, the degree of momentum refreshment, and the presence of a Metropolis correction recovers several familiar Hamiltonian and Langevin samplers as special cases. Three further, largely independent directions account for much of the remaining literature: modifying the metric, as in Riemannian HMC; modifying the Hamiltonian used by the acceptance test, as in modified or shadow HMC; and modifying the numerical integrator, as in multi-stage splitting schemes. We review these variants alongside the procedures used to tune and diagnose them, with particular attention to the effective sample size, whose standard estimators assume a reversibility that generalized and modified samplers lack, and ignore the importance weights that modified Hamiltonians introduce. We close with the open problems these methods face for hierarchical models, taking joint longitudinal and time-to-event models as the motivating example.
Arun K. Kuchibhotla, Sivaraman Balakrishnan, Larry Wasserman
Wed 18 Nov
stat.MEmath.STstat.MLstat.TH
pdf html
- interest Semiparametric inference and debiased machine learning 0.93
- reading nominal, procedures, linear models, inference
- favorites near “Simplifying debiased inference via automatic differentiation and probabilistic programming” 0.91
Abstract
We introduce COSI (COnfidence from Scale Invariance), a framework for constructing confidence sets for statistical functionals. COSI builds on the HulC (Hull-based Confidence), which uses median unbiasedness to construct confidence sets from batch estimates. COSI uses scale-invariance properties shared by many common estimators. We divide the data into batches, compute an estimate in each batch, and retain each candidate parameter value for which the batch estimates pass a test of scale invariance. The resulting confidence sets are asymptotically valid under weaker assumptions than classical Wald and bootstrap procedures require, and their coverage can be closer to the nominal level in small samples. Applications include robust inference for the mean, regression coefficients in misspecified linear models, tensor norms, and Fr\'echet means.
Cenchen Liu
Fri 13 Nov
econ.EMstat.ME
pdf html
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.87
- favorites near “Difference-in-Differences with multiple time periods” 0.89
- interest Causal inference and identification in econometrics 0.88
Abstract
Mean independence restrictions are widely used to identify average treatment effects, but the covariates used to formulate these restrictions are often observed or used in coarsened form. This paper shows that mean independence conditional on a finely measured covariate need not hold after the covariate is replaced by a coarsened version. Coarsening can therefore change the identifying content of the comparison rather than only the precision of estimation. I characterize the resulting coarsening bias for the ATT and ATE. For the ATT, the bias is driven by within-cell composition in untreated potential outcomes. For the ATE, an additional treatment effect heterogeneity term appears. I then introduce $c$-mean dependence, a nonparametric relaxation of exact mean independence at the coarsened level, and derive sharp bounds for average treatment effects. The parameter $c$ determines how far comparisons may extend across neighboring values of the coarsened covariate. I propose sample analogue estimators and bootstrap inference procedures for the resulting bound functions. In an application to voluntary military service using data from Angrist (1998), positive effects on early earnings, which primarily reflect military pay, hold even under the weakest form of $c$-mean dependence, whereas the evidence on later effects on civilian earnings is inconclusive once mean independence is relaxed.
Jamie Reason, Vik Shirvaikar, Mark van der Wilk, Chris Holmes
Tue 17 Nov
stat.MLcs.LG
pdf html
- reading incoherence, sequential, far, tell
- interest Reinforcement learning, policy learning and off-policy evaluation 0.88
- your work near “Efficient Online Estimation of Causal Effects by Deciding What to Observe” 0.86
Abstract
Many sequential decision-making procedures, such as active learning, implicitly assume that a model's current predictions agree with those it expects to make after updating on new data, a property we call coherence. Most models in use do not update by exact conditioning and are not guaranteed to be coherent, which can lead to worse decisions. Yet there is currently no general way to measure incoherence or to tell when it matters. We develop tools to measure incoherence and characterise when it affects decisions. We introduce the incoherence ratio, an MMD-based measure of how far any updating procedure departs from coherence, computed only from the procedure's own predictions and updates, and accompany it with an anytime-valid test of coherence. In Bayesian experimental design, acquisition objectives that are equivalent for coherent models choose different data under every approximate model we study, and the incoherence ratio tracks how far incoherence distorts them. We also introduce action coherence, which captures only the incoherence that changes a decision. What a resulting disagreement costs, however, is determined primarily by predictive accuracy. Sequential decisions depend on how a model updates as well as on what it predicts, so coherence, like accuracy, should be checked rather than assumed.
Suhwan Bong
Fri 13 Nov
stat.AP
pdf html
- reading welfare, policy, treatment, strategic
- interest Reinforcement learning, policy learning and off-policy evaluation 0.88
- your work near “Efficient Online Estimation of Causal Effects by Deciding What to Observe” 0.85
Abstract
The standard rule in unconstrained causal policy learning is simple: treat individuals with nonnegative CATEs. But this rule can fail when treatment eligibility is strategically manipulable. After a policy is announced, individuals may change the covariates used for assignment without changing their underlying treatment effects. We study how such responses affect treatment allocation and welfare. Using an access-burden representation, we characterize strategic treatment and decompose welfare loss into missed beneficial treatment and harmful treatment. For linear CATEs and quadratic manipulation costs, we derive closed-form threshold corrections. When manipulation costs are homogeneous, a single correction recovers the ideal positive-CATE allocation. With heterogeneous costs, welfare maximization, pointwise safety, and minimax group welfare regret can favor different thresholds. Simulations illustrate these tradeoffs and examine the effects of estimating treatment effects and behavioral responses.
Aarush Kulkarni, Khang Nguyen, William Chang, Yan Dai
Fri 13 Nov
cs.LGcs.AIstat.ML
pdf html
- favorites near “Double Reinforcement Learning for Efficient Off-Policy Evaluation in Markov Decision Processes” 0.90
- your work near “Timing as an Action: Learning When to Observe and Act” 0.86
- reading policy, feedback, regret, technique
Abstract
We study regret minimization in episodic Adversarial Markov Decision Processes (AMDPs) with unknown transitions and bandit feedback. In tabular AMDPs with $S$ states, $A$ actions, and horizon $H$, the best known $K$-episode regret is $\tilde O(HS\sqrt{AK})$, whose $\sqrt S$ gap to the $\Omega(H\sqrt{SAK})$ lower bound has long remained open. In linear AMDPs with $d$-dimensional features, after a five-year line of improvements starting from $K^{14/15}$, computationally efficient algorithms without a simulator or exploratory assumptions still only attain $\tilde O(K^{3/4})$ regret. A common challenge in both settings is to estimate the Q-function of a changing policy: past data comes from earlier policies that differ from (and are correlated with) the current one, while freezing the policy for future episodes worsens the dependency on $K$. We instead propose a general technique called introduced feedback delays: we estimate the Q-function of the policy using future episodes while still updating the policy, as if the feedback were delayed. While feedback delays are usually an obstacle imposed by the environment, here they are introduced voluntarily and, perhaps surprisingly, come at no extra cost. We develop Geometric Resampling with Rejection (GRR) to account for policy changes during the delay, and threshold the visitation probabilities to avoid long delays on rarely visited state-action pairs. With this unified technique, our policy optimization algorithms achieve, with high probability, $\tilde O(\text{poly}(H)\sqrt{SAK})$ regret in tabular AMDPs, resolving the open problem; and $\tilde O(\text{poly}(d,H)\sqrt K)$ regret in linear AMDPs, giving the first efficient algorithm optimal in $K$. They also naturally extend to delayed AMDPs with a total feedback delay $\mathfrak D$, yielding the first bounds optimal in $S$ and $\mathfrak D$ (tabular) and in $K$ and $\mathfrak D$ (linear).
Sara Wade
Mon 16 Nov
stat.ME
pdf html
- reading mcmc, carlo mcmc, authors, make
- favorites near “Variational Bayes under Model Misspecification” 0.93
- your work near “Valid Inference with Imperfect Synthetic Data” 0.82
Abstract
I congratulate the authors on their contribution, which addresses important challenges in understanding posterior distributions in Bayesian mixture models. As underlined by the authors, the Bayesian framework has several advantages, yet label-switching, identifiability issues, and the trans-dimensional nature of parameter spaces make exploiting the Bayesian approach and understanding the resulting posterior difficult in practice. The paper offers clear explanations and a simple, intuitive procedure to address this by clustering the Markov chain Monte Carlo (MCMC) draws of component-specific parameters. It is clearly illustrated on several examples.
M. Hashem Pesaran, Yimeng Xie
Tue 17 Nov
econ.EM
pdf html
- interest Causal inference and identification in econometrics 0.88
- reading standard errors, empirical, inference, literature
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.88
Abstract
This paper derives an exact expression for the bias of the linearized estimator of the variance for cluster-robust inference, which generalizes an earlier result derived by Bell and McCaffrey (2002). Both cases of continuous and binary treatments are discussed. Under the former it is shown that the bias of the linearized variance estimator and its jackknife variant are of order $G^{-1}$ relative to the true variance, where $G$ denotes the total number of clusters. A bias-corrected variance estimator (BCCV) is then proposed and shown to have a relative bias of order $G^{-2}$. In the case of models with binary treatments the orders of the bias of standard estimators and the proposed bias-corrected one are shown to be $\max (G_{1}^{-1},G_{0}^{-1})$ and $\max (G_{1}^{-2},G_{0}^{-2})$, where $G_{1}$ and $G_{0}=G-G_{1}$ are the numbers of treated and untreated clusters. Scale and degrees of freedom adjustments to the $t$-statistic computed using bias-corrected standard errors are also proposed. Small sample properties of the proposed BCCV test are investigated using Monte Carlo experiments. It is shown that the proposed test generally outperforms tests based on conventional and jackknife cluster standard errors, and performs comparably to wild bootstrap inference and the recent jackknife distribution-adjusted testing procedure by \cite{hansen2026jackknife}. The utility of the proposed test is illustrated\ using three empirical applications from the literature.
Mark Bech Knudsen, Helene Charlotte Wiese Rytgaard, Torben Martinussen
Fri 13 Nov
stat.ME
pdf html
- interest Causal inference and identification in econometrics 0.88
- reading multi-state, interventions, theory, causal
- favorites near “When do common time series estimands have nonparametric causal meaning?” 0.86
Abstract
We propose an identification theory for stochastic interventions on the transition dynamics in continuous-time multi-state models. Interventions are formalized as local modifications of a likelihood satisfying a causal invariance property. We show that identification from observed data is achieved when no unmeasured process simultaneously affects an intervened local characteristic and a competing or subsequent non-intervened local characteristic, covariate local characteristic, or censoring intensity. Our framework allows for intervening on both the timing and type of events, accommodates measured and latent covariates as time-varying marked point processes, and extends to right-censored data. We exemplify by considering identifiability in two concrete applications: one studying the effect of changing the timing of intrauterine insemination on the probability of pregnancy, and one studying the effect of changing the type of dysphagia relieving surgery on the risk of death in esophageal cancer patients.
Rodrigo Labouriau
Fri 13 Nov
stat.MEstat.CO
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.85
- interest Semiparametric inference and debiased machine learning 0.89
- favorites near “Simplifying debiased inference via automatic differentiation and probabilistic programming” 0.89
Abstract
Decomposable graphical models combine closed-form maximum likelihood estimation with local model comparison, which makes structure inference practical with many variables. The R package GraphInfer provides one likelihood framework for decomposable models of continuous, discrete and mixed data, the last as conditional Gaussian models. Structure inference ranges from exact minimum forests and stepwise search to stochastic global search. Nested models are compared by likelihood-ratio tests whose null distributions are exact in finite samples: conditionally on the discrete data for edges with a continuous endpoint, and conditionally on the discrete margins of the smaller model for edges between discrete variables. Missing values are handled by EM, by structural EM, which reuses complete-data searches through a weighted pseudo-sample of expected sufficient statistics, and by multiple imputation; selected models are fit to the observed-data likelihood. The package also treats the graphical models induced by irreversible event accumulation on trees: a search over the compatible graphs whose tree moves overcome an obstruction to single-edge moves, fitting of the full accumulation model with measurement noise and bootstrap tests of the accumulation hypothesis. GraphInfer is an independent, pure R successor to the archived package gRapHD. The implementation is validated against independent references, and simulation studies inform its defaults.
Dominik Ammon, Tobias Hartl, Rolf Tschernig
Wed 18 Nov
econ.EM
pdf html
- reading economics, theory, shocks, eigenvalues
- interest Macroeconometrics and time series 0.89
- favorites near “Advances in Nowcasting Economic Activity: Secular Trends, Large Shocks and New Data” 0.83
Abstract
Panels in economics, finance and climate science are often driven by common components that differ in memory, such as common trends, cycles, or joint shocks. We study approximate factor models whose common factors are fractionally integrated with heterogeneous and unknown memory and develop the asymptotic theory of principal components for the panel in levels. Our theory builds on the eigenvalues of the sample covariance matrix being ordered by memory asymptotically: a nonstationary factor group with memory $d>1/2$ generates eigenvalues of order $NT^{2d-1}$, whereas stationary factors generate eigenvalues of order $N$. Principal components estimate the factors consistently up to rotation, while adjacent factor groups are separated at a rate given by their memory gap. The information criteria of Bai and Ng remain consistent for the total number of factors, with their penalties acting as thresholds on eigenvalues that are asymptotically ordered by factor memory. When the nonstationary factors lose their variance advantage, estimation in levels remains consistent, whereas after differencing the factors can become undetectable. A Monte Carlo study shows that factor-number selection in levels tolerates more idiosyncratic noise than selection after first differencing. For monthly temperatures at $308$ European weather stations, our methods select two factors. The first is a persistent common warming factor. The estimated common component implies that 2021-2025 was $2.15^\circ\mathrm{C}$ warmer than 1960-1990, with the strongest warming in Eastern Europe. After first differencing, the criterion selects one factor, and loses regional heterogeneity as compared to the model in levels.
Trevor Sweeting, Stephen Jewson
Fri 13 Nov
stat.ME
pdf html
- interest Semiparametric inference and debiased machine learning 0.89
- your work near “Valid Inference with Imperfect Synthetic Data” 0.83
- reading theory, rise, regular, predictive distributions
Abstract
We construct objective priors for regular univariate parametric models designed to produce low coverage probability bias of Bayesian predictive distributions, giving rise to calibrated prediction in repeated use. This is achieved by defining the predictive discrepancy and deriving a class of minimum predictive discrepancy priors. In the multiparameter case these priors will often be data-dependent. The predictive efficacy of alternative global priors can be assessed by comparison with these priors. The effectiveness of the theory is illustrated by a number of examples.
Yacouba Boubacar Mainassara, Jean-Armel Bra, Landy Rabehasaina
Mon 16 Nov
stat.APstat.ME
pdf html
- favorites near “Priors for the Long Run” 0.92
- reading moment, empirical, procedures, data provide
- interest Macroeconometrics and time series 0.87
Abstract
In this paper we study the asymptotic properties of an autoregressive model of order p (AR for short), possibly non-centered, whose parameters switch according to an unobserved Markov chain. Departing from the conventional assumption of independent innovations, we merely require the errors to be uncorrelated but not necessarily independent, thereby widening substantially the range of potential applications of Markov-switching AR models. We establish sufficient conditions for the consistency and asymptotic normality of the proposed moment estimators. We also propose a consistent estimator of the asymptotic variance matrix of the moment estimators. The finite-sample performance of the proposed procedures is illustrated through Monte Carlo experiments and an empirical study based on hourly meteorological data, which together provide strong support for the theoretical findings.
Catalina B. García-García, Jorge L. Bazán Guzmán, Román Salmerón-Gómez
Tue 17 Nov
stat.ME
pdf html
- interest Semiparametric inference and debiased machine learning 0.91
- your work near “Valid Inference with Imperfect Synthetic Data” 0.83
- favorites near “Instrumental variable estimation in functional linear models” 0.89
Abstract
Ridge-type estimators are widely used to address multicollinearity in generalized linear models (GLMs), yet their empirical justification depends heavily on how multicollinearity is diagnosed. Following systematic review guidelines, we conduct a citation-based review of 120 real-data applications to examine current diagnostic practices. The findings reveal a lack of consensus across studies: approaches vary regarding data transformation prior to diagnosis, applying transformations to either the design matrix or the weighted information matrix, while others omit transformation entirely. Nevertheless, the condition number emerges as the most widely used diagnostic. We review existing methodologies for computing the condition number in GLMs and recommend the optimal approach, as it uniquely attains its minimum value under orthogonality while retaining the intercept. Finally, these formulations are implemented in R and used to replicate five real-data applications from the literature.
Hamza Ruzayqat, David M. Bortz, Vanja Dukić
Thu 19 Nov
stat.COstat.ME
pdf html
- reading equations, econometrics, trade, differential equations
- interest Computational methods for heterogeneous agent and dynamic economic models 0.83
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.84
Abstract
Weak-form methods have transformed inference for deterministic dynamical systems. By integrating the governing equations against smooth, compactly supported test functions, they trade noise-amplifying differentiation for noise-attenuating integration, yielding fast and robust estimates. Extending this idea to stochastic differential equations (SDEs) has proven harder, and a consistent single-trajectory formulation has remained out of reach. Applied naively in the time domain, the weak form makes the regressors and the errors functionals of the same noise, creating endogeneity that leaves weak-form least squares biased and inconsistent. We resolve it with instrumental variables, a classical idea from econometrics. Weak-form integrals over nearby disjoint time windows prove to be valid instruments, correlated with the regressors yet uncorrelated with the error, restoring consistency and asymptotic normality. The resulting generalized-method-of-moments estimator, WAIVES (Weak-form Asymptotically-unbiased Instrumental-Variable Estimation for SDEs), requires only a linear solve, with no likelihood evaluations, filtering, or simulation. It is therefore an order of magnitude faster than exact-likelihood methods where those exist, and it applies unchanged where they do not, with standard errors and calibrated confidence intervals supplied by the moment conditions. It also separates measurement noise from the intrinsic randomness of the dynamics. Across nine example systems, spanning linear, bistable, chaotic, excitable, state-dependent-diffusion, non-polynomial, hypoelliptic, and high-dimensional regimes (up to 120 dimensions), WAIVES recovers nonlinear drift and diffusion with negligible bias and near-nominal interval coverage. We close with two real-data illustrations: a 167-year North Atlantic sea-surface-temperature record and an optical-tweezer calibration.
Lorenzo Rimella, Augusto Fasano, Giovanni Rebaudo
Wed 18 Nov
stat.ME
pdf html
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.86
- interest Adaptive data collection and anytime-valid inference 0.88
- reading econometrics, online, encountered, inference
Abstract
The multinomial probit is a widely used model for categorical data in many applications, especially in econometrics and the discrete-choice literature. The computational challenges encountered when fitting this model still motivate ongoing research from the frequentist and Bayesian viewpoints. We consider a dynamic formulation based on a state-space model, where at each time one observes a sample from a multinomial probit whose parameter evolves according to Markovian dynamics, thereby inducing dependence across time. We show that the filtering and predictive distributions of the latent parameter belong to the unified skew-normal family, developing an associated i.i.d.\ sampler to approximate quantities of interest via Monte Carlo. Motivated by the computational bottlenecks of the sampler encountered for larger sample sizes, we also develop approximate methods for online inference based on assumed density filtering and expectation propagation. This gives more scalable, yet accurate, algorithms for online inference about the latent state and prediction of future observations. Finally, we derive a novel expectation maximization procedure to learn from the data the covariance matrix underlying the latent-utility representation of the multinomial probit. We investigate the performance of the proposed methods on simulated data and on a real dataset of reservations from an online booking platform.
Fabian Bleile, Sarah Lumpp, Mathias Drton
Thu 19 Nov
stat.MLcs.LGstat.ME
pdf html
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.85
- interest Causal inference and identification in econometrics 0.87
- reading causal, structural, equation, pairs
Abstract
Score-based structure learning estimates causal graphs by optimizing a data-dependent objective over structural equation models. We study how prior knowledge that selected pairs of variables are marginally independent can be incorporated into such causal discovery methods. Under faithfulness, marginally independent pairs cannot be connected by a trek, so the knowledge becomes a graphical no-trek constraint, which we call NOTREKS information. We introduce a power-series trek (PSTrek) penalty for gradient-based optimization and a trek constraint for order-based search, and we analyze why such information helps in finite samples and how it restricts the search space. Our numerical experiments illustrate to what extent NOTREKS information helps in causal discovery. In particular, using synthetic benchmarks with oracle NOTREKS information, we show that these constraints can improve both structural recovery and adjustment-based effect identification.
Zheyang Wu, Tiejun Tong
Tue 17 Nov
stat.COstat.ME
pdf html
- interest Semiparametric inference and debiased machine learning 0.90
- favorites near “Simplifying debiased inference via automatic differentiation and probabilistic programming” 0.89
- your work near “Valid Inference with Imperfect Synthetic Data” 0.82
Abstract
Calculating the moments, i.e., means, variances, and covariances, of the order statistics of a sample is critical, as they are foundations for order-statistics-based estimation and inference problems. In the early literature, the exact covariance matrices were tabulated only up to the sample size of 50 under normality. Recent software either provides exact calculation only for the means and variances of single order statistics, or obtains the covariances by approximations. We create the R package orderStatMoments to calculate the exact values of these moments for any ranks and any sample size. It covers thirteen built-in distribution families, specified by name and parameters as in R, and allows for continuous distributions defined by the user under relatively weak conditions. It also calculates the moments of any linear combination of order statistics, including sample quantiles, and Lloyd's best linear unbiased estimators with their variances. The work makes two contributions. First, the package provides the exact covariances between order statistics, which no software we found does. Second, we design methods based on the statistical properties of order statistics critical for ultra accurate computing at large sample sizes and various distributions. The accuracy is verified against closed forms at sample sizes up to $n = 10^6$, against the permutation identities on full matrices, and against 556 cells tabulated in seven publications, where it identified some errors far beyond the stated accuracy. A comparison with five R packages and Python's SciPy shows advantages of this package. Computing is efficient, with a time nearly independent of the sample size.
William T. Firmin
Fri 13 Nov
econ.EM
pdf html
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.85
- interest Causal inference and identification in econometrics 0.88
- favorites near “Difference-in-Differences with multiple time periods” 0.87
Abstract
In the regression discontinuity design, researchers assess the credibility of assumptions by testing the running variable's density for a discontinuity at the cutoff. This paper proposes a novel such test. The test uses q spacings on each side of the cutoff, where spacings are the widths between ordered observations. Spacings yield a continuous statistic that behaves well under weak assumptions. I study its properties in two asymptotic frameworks: one where q is fixed as the sample size increases and one where q slowly diverges to infinity. With q fixed, the test achieves exact asymptotic size control without user randomization. When q increases slowly, the test becomes consistent without sacrificing exact asymptotic size. Compared to the preferred version of Bugni and Canay's (2021) test, the only other test with comparably weak assumptions, this test reduces sensitivity to the tuning parameter and performs similarly when theirs is well-chosen.
Xiaoyu Li, Jiangxuan Long
Tue 17 Nov
cs.LGcs.CLstat.ML
pdf html
- interest Adaptive data collection and anytime-valid inference 0.88
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.84
- reading test-time scaling, pac-bayes, recipe, nominal
Abstract
Test-time scaling systems often choose how much to sample, which prompt to use, and which verifier or aggregation rule to trust after repeatedly inspecting a calibration set. Standard validation error bars are not valid after this adaptive selection. We model a test-time scaling recipe as a randomized predictor with a random cost and prove a time-uniform PAC-Bayes certificate that holds simultaneously over all calibration times, all posterior mixtures over recipes, and all data-dependent stopping rules. Thus the same event covers the recipe search that precedes deployment, not merely the fixed recipe finally reported. For a finite grid of budgets and verifiers, the penalty is the familiar $\log M/n$, which in the worst case no certificate valid after selection can avoid, while non-uniform priors and localized posterior mixtures can give sharper certificates for structured searches. We specialize the result to self-consistency and show a complementary margin law: plurality voting improves exponentially in the sample budget only on examples where the correct answer is the generator's unique mode; otherwise more compute can provably amplify errors. The same margin analysis yields an oracle water-filling law for compute allocation. A synthetic audit quantifies the optimism caused by adaptive budget selection, and a 192-example GSM8K audit using gpt-4.1-mini generations illustrates risk-cost reporting for a real inference pipeline. At scale, with two models on GSM8K and MATH-500, selected-best error bars fall below the out-of-split GSM8K risk in up to 22.5% of random splits at a nominal 5% while the certificate never does, and on the full GSM8K test split it tightens to 0.109 (held-out risk 0.039). The results give a lightweight, distribution-free way to report risk and compute claims for adaptive inference systems.
Paul Schneider
Wed 18 Nov
math.STq-fin.STstat.TH
pdf html
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.83
- interest Causal inference and identification in econometrics 0.87
- favorites near “Measuring Price Selection in Microdata: It's Not There” 0.84
Abstract
We estimate the probability that an item with given covariates belongs to a set when the cross-section changes over time and items and periods are dependent. We propose two convex models whose fitted probabilities are nonnegative by construction and remain valid whatever the dependence between items, for which we establish consistency, finite-sample bounds and limit distributions for serially dependent panels. One suits events of moderate or high probability, the other rare events, and half the disagreement between their fits estimates a lower bound on the misspecification error of the worse fit, without knowledge of the true probabilities. On monthly US stock failures, the rare-event model predicts better than the ordinary logit and the classical failure logit. With the eight classical predictors, most of the gain comes from basis functions of the rank angle of each predictor and summary.
Jeffrey Tse
Tue 17 Nov
econ.EMstat.ME
pdf html
- favorites near “When do common time series estimands have nonparametric causal meaning?” 0.88
- interest Causal inference and identification in econometrics 0.89
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.83
Abstract
We compare three estimators in a two-sample summary-data instrumental variables (IV) model: inverse-variance weighting (IVW), debiased inverse-variance weighting (dIVW), and the profile score (PS) estimator. The framework assumes independent samples and instrument-specific summary estimates, with two-sample Mendelian randomisation as a leading application. We compare existing fixed- and many-weak-instrument results and the sufficient conditions for the three estimators. Under the variance scaling used in the profile-score literature, the principal collective-strength conditions for PS and dIVW are equivalent, although their additional restrictions on instrument concentration and growth differ. We then compare their asymptotic variance formulae for fixed and diverging numbers of instruments. For fixed $L$, IVW and dIVW have the same first-order asymptotic variance. Under a common variance-ratio condition, the PS variance formula exceeds the IVW variance by a positive term of order $n^{-2}$, compared with the order-$n^{-1}$ leading variance. For diverging $L$, we establish variance orders across instrument-strength regimes and the exact ordering $\mathsf{V}_{\mathtt{IVW}} < V_{\mathtt{PS}} \leq \mathsf{V}_{\mathtt{dIVW}}$ under the same variance-ratio condition. Absolute variance gaps vanish when $\sqrt{L}\kappa\to\infty$, whereas relative equivalence holds if and only if $\kappa\to\infty$. These comparisons concern asymptotic variance formulae rather than finite-sample efficiency or mean squared error.
Wen Zhang, Yanyuan Ma, Liping Zhu
Thu 19 Nov
stat.ME
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.85
- reading nuisance, procedures, question, natural
- interest Causal inference and identification in econometrics 0.85
Abstract
In many studies, basic covariates are routinely recorded at negligible cost, whereas acquiring additional covariates requires expensive or time-consuming procedures. Thus, a natural question is whether it is worth to collecting these additional covariates. Using the reduction in prediction interval width as criterion, we evaluate and test the gain brought by the additional covariates and hence answer this question. We consider the gain both marginally and conditionally on the observable characteristics. The conditional formulation captures subpopulation heterogeneity and supports individualized data acquisition decisions. To avoid restrictive parametric assumptions, we adopt a model-free framework that accommodates flexible nonparametric or machine learning methods. We propose an efficient influence function based method, with proven theoretical properties to achieve our purpose under mild nuisance convergence rates. Interestingly, we find the natural conformal prediction based method cannot achieve the same purpose: It fails in the marginal case and imposes much more stringent or even infeasible requirement on the nuisance parameter estimation procedures. We demonstrate the superior finite-sample performance of the proposed procedures through simulations and real data applications.
Matthew Zurek, Yudong Chen
Tue 17 Nov
cs.LGcs.ITmath.ITmath.OCmath.PRstat.ML
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.85
- favorites near “Double Reinforcement Learning for Efficient Off-Policy Evaluation in Markov Decision Processes” 0.88
- reading bracketing, policy, empirical, state
Abstract
We study the sample complexity of the empirical planning or plug-in approach for reinforcement learning (RL) with a generative model. We show the discounted plug-in approach finds an $\varepsilon$-optimal policy with $O(SA (1-\gamma)^{-3} \varepsilon^{-2} \log(SA/\delta))$ samples, where $S, A, \gamma, \delta$ are the state and action sizes, discount factor, and failure probability, respectively. In the average-reward setting, we show that discounted plug-in and direct (undiscounted) plug-in achieve $O(SAH\varepsilon^{-2}\log(SA/\delta))$ and $O(SAD\varepsilon^{-2}\log(SA/\delta))$ complexities, respectively, where $H$ upper-bounds the optimal bias span and $D$ is the diameter. Combining with our lower bounds for nonadaptive generative models, these results resolve the minimax sample complexity of discounted RL and average-reward RL --- in both the span- and diameter-based classes --- up to constant factors. Our main technical contribution is a novel Bellman residual bracketing technique which improves the efficiency of prior absorbing-MDP-based concentration analyses.
Daisuke Matsuno, Kanta Naito
Tue 17 Nov
stat.ME
pdf html
- interest Semiparametric inference and debiased machine learning 0.90
- favorites near “Orthogonal Statistical Learning” 0.90
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.82
Abstract
Maintaining both interpretability and accuracy remains a fundamental challenge in regression analysis. This paper addresses these two objectives through a semiparametric approach that combines a parametric crude guess with its nonparametric adjustment. In particular, the nonparametric adjustment is implemented via a machine learning technique called ``stagewise empirical risk minimization algorithm.'' Once a set (dictionary) of functions (words) is prepared, this algorithm constructs a convex combination of words selected iteratively from the dictionary. This selection is based on the empirical risk derived from the Bregman divergence, which enables the proposed estimators to be applied in broad settings, including generalized linear models. This paper examines these proposed semiparametric estimators theoretically and numerically. In the theoretical study, we derive a non-asymptotic error bound for the estimators constructed from a general data-dependent dictionary. Among the concrete dictionaries proposed in this paper, some can be utilized to correct variable selection in LASSO-based parametric regression, which is investigated in the numerical study. Moreover, simulation studies investigate practical performance of the proposed estimators, while real data applications further confirm their predictive performance. These studies indicate that the proposed semiparametric estimators can address the above challenge in several settings.
Nathaniel Xu, Feng Liu, Danica J. Sutherland
Fri 13 Nov
stat.MLcs.LG
pdf html
- reading wealth, sequential, historically, game
- interest Adaptive data collection and anytime-valid inference 0.89
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.83
Abstract
Sequential testing avoids some of the many controversies and drawbacks related to traditional batch testing with p-values. In particular, the testing by betting paradigm has developed rapidly in recent years; it frames the level of evidence against the null as the wealth of a player in a betting game. Naturally, how much a player should bet is a key consideration, and should reflect their confidence in what they know about the next observed data point. Much current work uses standard betting strategies developed for adversarial problems. In the stochastic setting, where the observed data is independent and stationary, these approaches may be suboptimal. We propose a simple betting strategy, follow the leave-one-out e-power (FLOE), that exploits the i.i.d. nature of a data stream to estimate the e-power of our current test on the underlying distribution. We average not over the historical evidence collected so far, as generic online learning-based methods do, but instead over the equally-likely possible data sequences we might have seen based on shuffling the data. With an additional principled regularization technique, this simple drop-in method yields substantial improvements in testing performance on a variety of kernel two-sample and independence testing problems.
Matteo Barigozzi, Enzo D'Innocenzo
Tue 17 Nov
econ.EMstat.ME
pdf html
- interest Macroeconometrics and time series 0.90
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.85
- favorites near “Advances in Nowcasting Economic Activity: Secular Trends, Large Shocks and New Data” 0.83
Abstract
We propose a factor model for high-dimensional time series where the dynamics of the latent factors is generated by a multivariate score-driven model, thus allowing to model nonlinearities and heavy tails in both the conditional mean and the conditional variance. Estimation is in two steps: first the factors are extracted via Principal Components and then the parameters of the score-driven model are estimated via Quasi Maximum Likelihood based on the estimated factors. Since factors are not identified, we first establish set-consistency and asymptotic normality for the estimated parameters governing their dynamics, as both the number of time series and the sample size diverge to infinity. Then, we construct valid asymptotic prediction intervals, which do not depend on the chosen identifying constraints for the factors. Numerical results confirm the goodness of our estimator.
Seth Flaxman, Jess Carr, Annabel Jakob, Marek Masiak, Dino Sejdinovic
Fri 13 Nov
cs.LGstat.ML
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.86
- reading causal, observational, question, methods causal
- favorites near “Using Imperfect Surrogates for Downstream Inference: Design-based Supervised Learning for Social Science Applications of Large Language Models” 0.85
Abstract
In regression and classification tasks with tabular datasets, most of the literature on explanation in ML (e.g. SHAP, permutation importance) has focused on predictive attribution: does feature $X_i$ help predict outcome $Y$? At its heart, predictive attribution captures correlation, not causation. We instead ask a causal question: would manipulating feature $X_i$ change outcome $Y$? We define a natural quantity to answer this question, the feature intervention effect: $\Delta_i = E[Y\mid do(X_i=+1)]-E[Y\mid do(X_i=-1)]$. In general, estimating this quantity from observational data alone would entail strong causal assumptions. We propose FADO (Feature-wise Amortized Do-Operators), an amortized inference method using prior-fitted networks trained on synthetic tabular datasets generated by structural causal models. During training, the synthetic data-generating process is known, so we can compute ground-truth intervention effects $\Delta_i$, and train a model to predict them at test time from a purely observational tabular dataset. As we demonstrate through extensive experiments, our new causal explainability method is easy to train and provides better explanations under a range of settings for which standard predictive attribution fails, including proxy variables, confounders, colliders, and placebos. We evaluate it on real, synthetic and semi-synthetic datasets, with and without hidden confounders, and compare it to existing predictive attribution methods and state-of-the-art causal inference methods.
Chengyu Cui, Gongjun Xu
Wed 18 Nov
math.STstat.TH
pdf html
- favorites near “Variational Bayes under Model Misspecification” 0.91
- reading latent distributions, latent distribution, latent variable, advances
- interest Semiparametric inference and debiased machine learning 0.87
Abstract
Latent variable models are popularly used to characterize dependence among many observed variables through underlying latent variables. Recent advances in statistics and machine learning methods have encouraged growing interest in studying latent variable models with flexible latent distributions that capture complex latent structures. This flexibility, however, raises a fundamental question of whether model parameters can be identified when the latent distribution is unknown and must itself be learned. In this paper, we show that exact identifiability fails almost everywhere, in the sense that distinct but arbitrarily close pairs of model parameters and latent distributions can yield exactly the same distribution for the observed variables. Nonetheless, the resulting identification ambiguity is small. In particular, we prove that the identification ambiguity, measured by the diameter of the set of parameters inducing the same marginal distribution, vanishes at a superpolynomial rate of $\exp\{-c(\log p)^2\}$ as the number of observed variables $p$ grows. Building on this result, we show that the failure of exact identifiability need not preclude consistent estimation or valid inference of the model parameters and latent distribution. These results suggest that, when many variables are observed, the identifiability issue is far less of an obstacle to estimation and inference in latent variable models than is commonly expected.
Guillaume Baquet, Germain Van Bever
Tue 17 Nov
math.STstat.MLstat.TH
pdf html
- reading hilbert, hilbert space, operator, ill-posed
- interest Semiparametric inference and debiased machine learning 0.91
- favorites near “When do common time series estimands have nonparametric causal meaning?” 0.82
Abstract
We study nonparametric instrumental variable regression when the outcome takes values in a separable Hilbert space $\mathbb{Y}$, for instance a curve, a density or the kernel mean embedding of a distribution. The structural function solves an ill-posed operator equation whose conditional expectation operator need not be injective, and whose solutions need not belong to the hypothesis space. We analyse a two-stage kernel estimator that extends kernel instrumental variable regression to $\mathbb{Y}$-valued outcomes and allows spectral regularisation in both stages, including Landweber iteration and Showalter's method in the second stage. We show that the second stage is a regularisation method in the Hilbert scale generated by the inverse of the kernel integral operator. This viewpoint identifies the solution towards which the estimator converges: the minimum-norm solution in the reproducing kernel Hilbert space in the well-specified case and, in the misspecified case, the unique solution in the range of the continuous extension of the projection operator associated to the well-specified setting. Under source conditions and link conditions expressed through general index functions, we derive non-asymptotic bounds in a scale of interpolation norms that hold with high probability in terms of the first-stage error, with or without identification and in the well- and misspecified settings. For polynomial link conditions, we recover the rate of scalar kernel instrumental regression and extend it to targets outside the hypothesis space; for exponential link conditions, we obtain logarithmic rates in the severely ill-posed regime.
Dennis Kristensen, Ruolei Zhu
Fri 13 Nov
econ.EMstat.ME
pdf html
- reading rational, econometric, choice, primitives
- interest Reinforcement learning, policy learning and off-policy evaluation 0.85
- favorites near “When do common time series estimands have nonparametric causal meaning?” 0.84
Abstract
This paper introduces a novel class of econometric models, called Rational Inattention-Additive Random Utility Models. It is based on the discrete choice rational inattention model of Matvejka and McKay (2015) with additive unobserved heterogeneity incorporated. We show that some of the systematic components of the primitives of the model are nonparametrically identified even if the distributions of unobserved variables and the prior of decision makers are both unknown to the Econometrician. If additional information about the prior is known to the Econometrician, even stronger identification results become available.
Ayya Alieva, Gautam Iyer, Alexei Novikov, Lenya Ryzhik
Tue 17 Nov
math.STmath.PRstat.TH
pdf html
- reading manifold, continuous time, data manifold, argument
- favorites near “Eigendecompositions of Transfer Operators in Reproducing Kernel Hilbert Spaces” 0.86
- interest Kernels, operators and functional data 0.85
Abstract
We study diffusion models under the manifold hypothesis with possibly large score estimation and discretization errors, and show that while they may not exactly recover the data distribution, they will exactly recover the support (or data manifold). In continuous time, we use a Lyapunov function to show that even if the relative score estimation error is fixed (and not necessarily vanishing), then the rate at which the process approaches the support is $O(\sqrt{(T-t)|\log(T-t)|})$ as $t \to T$, where $T$ is the final time. In discrete time, we use geometric time steps with a fixed (non-vanishing) ratio and show that with a fixed (also not small) relative score estimation error, the process approaches the data manifold exponentially with the number of iterations. The proof relies on controlling a Gaussian measure of a neighborhood of the cut-locus of the data manifold, using a polynomial approximation argument combined with quantitative semi-algebraic estimates.
Yongsik Sim
Wed 18 Nov
stat.ME
pdf html
- favorites near “Priors for the Long Run” 0.86
- interest Macroeconometrics and time series 0.88
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.82
Abstract
Autocorrelation-based timescale estimation often fits a single exponential to a finite sample autocorrelation vector, even when the underlying process is not autoregressive of order one. We examine the first-order uncertainty of this fixed-lag projection under model misspecification. Standard extremum-estimation arguments yield a sandwich variance whose denominator is the full population curvature, including a residual-second-derivative term omitted by a Gauss-Newton approximation. A conditional sign rule and two stationary Gaussian examples with exact standard-error ratios 14/13 and 13/15 show that omission can underestimate or overestimate uncertainty. Full-to-Gauss-Newton population standard-error ratios are 1.34 and 1.45 in two K=36 noise-mixture models and range from 0.38 to 1.37 across an evaluated published-process parameter grid. In a follow-up experiment with 24,000 synthetic series, empirical standard deviations in four Gaussian models are consistent with the full-curvature scale at the main sample length, whereas the Gauss-Newton scale is substantially too small in the two noise-mixture models. In all 4,000 observed replicates of the primary condition, full-curvature intervals contained their Gauss-Newton counterparts, constraining the observed coverage gain to be nonnegative. Fixed-bandwidth HAC targets a smoothed long-run variance, and full sample curvature can worsen finite-sample coverage in an exactly specified model. Bandwidth-specific results, directional coverage errors, and oracle comparisons separate curvature effects from long-run variance estimation and transformation effects. These findings identify an application-specific discrepancy in a published autocorrelation-domain variance formula and characterize its asymptotic and finite-sample consequences.
Marc Braun, Adel Daoud, Jose M. Peña
Wed 18 Nov
stat.MLcs.LG
pdf html
- interest Causal inference and identification in econometrics 0.86
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.86
- reading counterfactual, tell, consistent, causal
Abstract
In general, counterfactuals are not identified from data alone, because many causal models produce the same observed distributions but different counterfactuals. A common approach to circumvent this problem is to restrict the mechanism to a class in which it becomes unique, such as triangular monotone maps or optimal transport maps. However, whether the assumed class is correct is usually not examined. In this work, we study what the data can tell us when covariates are available that influence the outcome only through its latent cause. First, we characterise all counterfactual maps that are consistent with the data. They differ from the true map by a group of symmetries that preserve the conditional outcome distributions given the covariates. The more these distributions vary across covariate values, the smaller the group. Second, we show that for the classes above, consistency with the data is equivalent to a single conditional independence, which yields a test of the assumed class. We illustrate both results on synthetic data and on the ACTG 175 HIV trial.
Gérard Biau
Tue 17 Nov
math.STstat.TH
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.85
- interest Adaptive data collection and anytime-valid inference 0.89
- favorites near “Simplifying debiased inference via automatic differentiation and probabilistic programming” 0.86
Abstract
Modern data-analysis pipelines increasingly involve searching among candidate analyses before reporting an estimate. We investigate the replicability of the resulting conclusions by studying independent repetitions of the entire search-and-report procedure on new samples. Can the reported estimates become increasingly similar across studies while remaining systematically displaced from the scientific target? We first show that this can happen in a simple Gaussian model. Although every prespecified candidate estimator is unbiased and consistent, searching over exponentially many candidates within each study yields a selected estimator converging to a limit different from the target, a selection-induced pseudotruth. Across independent repetitions, the reported estimates thus become replicable but systematically inaccurate, while the selected candidates differ with probability tending to one. We then study dependent Gaussian candidates, including marginal regression coefficients, and show that Gaussian width governs the displacement, which persists when squared width is of the same order as sample size. Metric entropy and generic chaining reveal that the effective complexity of the search is determined by the geometry of statistically distinguishable analyses rather than by their nominal number. Large-deviation arguments further extend the phenomenon beyond Gaussian models. Numerical agreement between independent studies therefore need not indicate either a stable analytical path or proximity to the scientific target.
Hanqi Wang
Mon 16 Nov
econ.EMstat.ME
pdf html
- reading rational, semiparametric, choice, wage
- interest Reinforcement learning, policy learning and off-policy evaluation 0.86
- favorites near “Price Level and Inflation Dynamics in Heterogeneous Agent Economies” 0.79
Abstract
People's choices depend on both the payoffs of the available options and what they learn about those payoffs before deciding. This paper shows how choice data identify both unobserved payoff heterogeneity and the cost of information acquisition in a rational inattention discrete-choice model. I establish conditions for semiparametric point identification using exogenous payoff and information-cost shifters, without imposing parametric forms on the taste-shock distribution or on how the unit cost of information varies with observed characteristics. In the model with binary choice, I construct an unobserved choice threshold whose cumulative distribution function is the choice probability as a function of the payoff shifter. Moments of this threshold identify the payoff coefficients and the unit cost of information at each value of the information-cost shifter. A deconvolution argument then identifies the taste-shock distribution. I then extend identification to the model with multinomial choice. To estimate the model, I propose a sieve minimum-distance estimator based on observed market shares and establish its consistency. For inference, I construct confidence regions by test inversion because binding parameter constraints can make the estimator's limiting distribution nonnormal. In an application to Amazon Mechanical Turk, model counterfactuals show that lower unit costs of information raise the average task-taking probability and reduce wage elasticity.
Callum Wu, Jim E. Griffin
Tue 17 Nov
stat.ME
pdf html
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.85
- reading rational, log, gains, recalibration
- interest Macroeconometrics and time series 0.86
Abstract
Combined density forecasts are often miscalibrated, and the miscalibration can change over time. We propose a time-varying framework that recalibrates any combined forecast through applying a monotone map to its probability integral transform. We consider a Beta CDF or a rational quadratic spline as the map, and their parameters are updated either via score-driven filters combined with dynamic model averaging, or by a convolution-based update that additionally conditions on features of the forecast distribution. In simulations, the spline map corrects multimodal miscalibration that the Beta map cannot, and the convolution update achieves the highest log score when the miscalibration switches between regimes. We apply the method to nine G10 exchange rates and US inflation, recalibrating forecast distributions from both a loss-discounting linear pool and Bayesian predictive synthesis. In both applications, recalibration improves PIT uniformity, with gains in forecast accuracy most evident when the original forecasts are miscalibrated.
Florian Gunsilius, David Van Dijcke, Kaspar Wüthrich
Wed 18 Nov
stat.MEcs.AIecon.EM
pdf html
- reading empirical, science, nominal, inference
- interest Adaptive data collection and anytime-valid inference 0.86
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.81
Abstract
Empirical research is changing rapidly as AI coding agents are becoming increasingly proficient in every step of it, from writing code to manipulating data. We argue that while the potential of coding agents for science is vast, their use in hypothesis testing poses a substantial threat to classical inference, as these agents enable a supercharged form of selective inference. In particular, we argue that coding agents carrying out empirical research introduce a new type of post-selection inference that combines human-like, open-ended search over an implicit space of estimators with programmatic replayability of both the selection and the estimation process. Indeed, we find that on simulated data with zero effects, four frontier agents report significant p-values between 54% and 98% of the time when asked to produce a compelling finding. We propose a simple procedure for detecting and correcting these distortions. After an agent has completed its analysis, we reset and replay it on subsamples of the data to estimate the agent's relative probability of selecting that analysis when it looks statistically favorable. For agents with stable selection behavior, the resulting corrected test is asymptotically valid and consistent conditional on selection. Importantly, the correction is learned from the black-box agent's actual choices rather than a model of how it searches, so it remains valid even as AI technology continues to evolve rapidly. In simulations, the correction robustly lowers the high false rejection rates to near the nominal 5%. We illustrate the method on applications from two published papers.
Taole Sha, Shibo Yu, Edwin Fong
Wed 18 Nov
stat.MEmath.STstat.TH
pdf html
- favorites near “Variational Bayes under Model Misspecification” 0.91
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.83
- interest Bayesian computation and structural estimation 0.86
Abstract
Bayesian nonparametric density estimation is challenging in high dimensions, where traditional approaches like the Dirichlet process (DP) mixture rely on computationally intensive Markov chain Monte Carlo or approximations. We introduce the partition posterior bootstrap (PPB), where a flexible prior is constructed by smoothing a DP over a partition of the sample space. Conditioned on this partition, exact posterior sampling reduces to a simple Bayesian bootstrap-like algorithm. Importantly, we select the partition using a greedy empirical Bayes (GEB) algorithm which scales to high dimensions and large sample sizes. GEB greedily maximizes the closed-form marginal likelihood via local split decisions arising naturally from its factorization, yielding an adaptive binary tree partition. Our key theoretical result is that the GEB tree belongs to an admissible family with high probability, splitting all identifiable signals while avoiding over-splitting noise. As a result, the PPB posterior with the GEB tree attains adaptive contraction rates. Our theory holds under weak, verifiable assumptions, which we demonstrate in numerous examples. Empirically, PPB delivers accurate density estimates and uncertainty quantification at a fraction of the computational cost of other approaches, making it a practical off-the-shelf choice for high-dimensional Bayesian density estimation with theoretical guarantees.
Paul Navas
Mon 16 Nov
math.STstat.MEstat.TH
pdf html
- reading consistent, equations, connection, functions
- interest Semiparametric inference and debiased machine learning 0.88
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.83
Abstract
On Euclidean space, the losses that produce consistent M-estimators of a functional are its strictly consistent scoring functions (Dimitriadis, Fissler and Ziegel, 2024), and Osband's principle characterizes these through identification functions, which define method-of-moments estimators. We extend this characterization to the probability simplex. We define the residual, which on Euclidean space is a subtraction, as the inverse exponential map of an affine connection. It identifies the exponential barycenter of the connection: the arithmetic, normalized geometric and Fr\'echet means for the mixture, exponential and Levi-Civita connections. Its integral along geodesics is the canonical divergence of Ay and Amari (2015). A geometric form of Osband's principle decides which losses each connection admits. For the flat connections every strictly convex generator yields a consistent Bregman loss. For the curved $\alpha$-connections, half the squared Fisher-Rao distance is (up to scale) the only consistent loss for the Fr\'echet mean, and for $\alpha \notin \{-1, 0, 1\}$ no such loss exists and only the method of moments remains. The residual is integrable only at $\alpha = 0$, sharpening the integrability claim of Ay and Amari. The characterization yields the estimating equations of regression on the simplex and the calibration decomposition of compositional forecasts.
Ramesh Johari, Anushka Murthy
Fri 13 Nov
stat.MEecon.EMecon.TH
pdf html
- reading randomized, preferences, nominal, larger
- your work near “Efficient Online Estimation of Causal Effects by Deciding What to Observe” 0.84
- interest Adaptive data collection and anytime-valid inference 0.85
Abstract
Two-sided marketplace platforms often use randomized experiments to evaluate new features, raising the question of whether to randomize customers or listings. Randomizing the smaller (\emph{short}) side can reduce interference bias due to weaker competition, while randomizing the larger (\emph{long}) side provides a larger nominal sample, potentially reducing variance. On the surface, these competing considerations suggest a bias-variance trade-off in choosing which side to randomize. We investigate this trade-off in a two-sided marketplace with first-come, first-served (FCFS) booking and heterogeneous customer preferences, studying the bias and variance of difference-in-means estimators of the global average treatment effect. Our findings suggest that rather than a bias-variance tradeoff, in fact short-side randomization is generally preferable on {\em both} dimensions. We prove these results in an asymptotic regime where market imbalance increases. We establish conditions under which (1) short-side randomization eliminates first-order interference bias, while long-side randomization retains it; and (2) short-side randomization also achieves a variance rate of no larger asymptotic order than long-side randomization. The key insight is that although the long side contains more participants, the number of matches is constrained by the population on the short side, limiting the apparent variance advantage of randomizing a larger sample. Our conditions accommodate heterogeneous preferences and treatment effects and are satisfied by familiar logit models of customer choice. Together, these findings support randomizing on the short side, despite the larger sample size available on the long side.
Chen Cheng
Tue 17 Nov
math.STmath.PRstat.MLstat.TH
pdf html
- reading hilbert space, theory, hilbert, functional data
- interest Kernels, operators and functional data 0.89
- favorites near “Eigendecompositions of Transfer Operators in Reproducing Kernel Hilbert Spaces” 0.88
Abstract
Spectral properties of the sample covariance are key to modern machine learning with high-dimensional data. Random matrix theory (RMT) supplies powerful tools; however, most classical results assume proportional asymptotics, where $d/n$ tends to a finite limit. In functional data analysis and kernel methods, a fixed infinite-dimensional Hilbert space is often the more natural setting for the observations, beyond the scope of proportional asymptotics. This calls for a dimension-free spectral theory for i.i.d. observations $\boldsymbol{x}_1, \dots, \boldsymbol{x}_n$ in a Hilbert space $\mathcal{H}$ with trace-class population covariance $\Sigma$. Building on the dimension-free ridge regression analysis of Cheng and Montanari (2024), we establish two deterministic equivalents with explicit nonasymptotic error bounds. The first approximates the empirical spectral law $\widehat{\mu}_n$ of the sample Gram matrix by a deterministic law $\mu_n$. The second characterizes the conditional prediction risk of analytic spectral filters in linear regression. Previous sharp approximations to prediction risk in this setting have focused on ridge regression and closely-related problems. Our analysis covers a broader class of filters at the bulk scale, with an explicit user-friendly spectral stability condition to check. We apply our theory to power and critical spectral laws, and we recover the proportional regime. For prediction, we study (negative) ridge estimators, iterated ridge, and gradient flow. Mathematically, building on the idea of stochastic interpolation, we construct a deterministic backward path to interpolate the spectral parameter. The resulting bounds on resolvent and mixed-resolvent traces give mesoscopic approximations to the spectral law within the bulk and, by Cauchy's formula, deterministic equivalents for the prediction variance and squared bias of analytic spectral filters.
Mohamed Chaouch
Tue 17 Nov
stat.MEcs.LGstat.ML
pdf html
- favorites near “Priors for the Long Run” 0.86
- interest Macroeconometrics and time series 0.88
- reading functional, forecasting, nominal, curves
Abstract
Forecasting a curve, rather than a handful of scalar summaries of it, requires a notion of uncertainty that is itself functional. We propose FARCH, a functional nonlinear autoregressive model with conditional heteroscedasticity, in which the conditional mean and the conditional volatility of the next curve are forecasted jointly and nonparametrically, by reusing a single kernel weighting scheme --- built on a wavelet semi-metric between curves --- first on the curves themselves and then on their squared residuals. From this pair of estimators we derive two constructions of prediction bands: a pointwise band, obtained from marginal scenario quantiles, and a uniform, sup-norm band, calibrated for decisions that concern the entire forecasted path rather than any single instant. Applied to day-ahead forecasting of half-hourly French national electricity demand (1996--2009), and compared, under an identical protocol, against four existing prediction-interval constructions that do not model the conditional variance explicitly, the uniform FARCH band attains pointwise coverage statistically indistinguishable from its nominal target at the 80%, 90% and 95% levels, and a curve-wise coverage of 90.1%, significantly above the 70.8% reached by the best competing method, at the cost of a wider interval. Pointwise bands, in contrast, undercover regardless of whether the variance is modelled explicitly, indicating that state-dependent volatility and simultaneous calibration are two distinct ingredients that a trustworthy prediction band must combine.
Ulrich Hounyo
Fri 13 Nov
econ.EMstat.ME
pdf html
- reading institutions, econometric, income, partially identified
- your work near “Perturbation Methods for Incomplete Markets Economies” 0.85
- favorites near “SVAR (Mis)Identification and the Real Effects of Monetary Policy Shocks” 0.79
Abstract
When data reject an econometric model, the researcher must decide which identifying assumptions to give up. If several equally preferred repairs restore compatibility, the repair itself is unknown, and inference under any one of them can exclude structural values supported by the others, even when that inference is exact. I propose Post-Relaxation Projection (PR), which carries one simultaneous confidence event through every repair that may be minimal. With repair-by-repair primitive bounds, PR equals the union of all post-relaxation targets compatible with the band, so no smaller band-based set has the same guarantee. I derive limits on learning the repair, the exact perturbation radius of the minimal-repair correspondence, oracle-equivalent adaptation, and a reduced-form version valid at zero first stages. In cross-country data on institutions and income, the data do not settle which historical instrument should be dropped, yet every admissible minimum repair implies a positive effect of institutions.
Rahul Agarwal, Ankit Agarwal
Tue 17 Nov
stat.MLcs.LG
pdf html
- reading equations, band-limited, fit, nonparametric
- favorites near “Fit without fear: remarkable mathematical phenomena of deep learning through the prism of interpolation” 0.84
- interest Kernels, operators and functional data 0.85
Abstract
Band-limited maximum likelihood (BLML) estimates a density as the square of a band-limited function. The estimate is non-negative, and when the square root of the true density is band-limited its error falls at the parametric rate $n^{-1}$. Its algorithm bins the data, which is fast only in one or two dimensions: from three on every point has its own bin, the kernel system is dense, and memory grows as $n^2$, 40 GB at $10^5$ points. Whether the rate survives in higher dimension, where kernel estimators slow down, had not been tested. We study dense data, densities that fill their space, not data near a low-dimensional manifold. We show that the binned equations are the stationarity condition of a strictly convex objective and minimise it by block coordinate descent without forming the kernel matrix: memory is linear in $n$, a sweep costs $O(n^2 d)$ and a fit takes about ten, and one A100 GPU fits $10^6$ points in eight dimensions in about ten minutes. On band-limited truths the error then decays as $n^{-1}$ in every dimension we tested, up to eight, while the slopes of kernel estimators flatten from -0.67 to -0.28, and it is 2 to 18 times lower than a neural spline flow's. A streaming fit learns the cut-off, BLML's only tuning parameter, from analytic gradients of the held-out likelihood, for about the cost of one more fit. On real data BLML is the strongest nonparametric estimator we tested: it leads kernel and nearest-neighbour estimators on electricity prices by 0.27 to 0.47 nats per hour, and a tuned kernel estimator on the full POWER and GAS benchmarks by 0.39 and 1.58 nats per point, placing it between early and recent normalizing flows. It loses on densities with a kink.
Jason Beh, Jérôme Morio
Tue 17 Nov
stat.COmath.STstat.MEstat.TH
pdf html
- reading nominal, expectations, density, functions
- favorites near “Simplifying debiased inference via automatic differentiation and probabilistic programming” 0.88
- interest Adaptive data collection and anytime-valid inference 0.86
Abstract
Simulation studies often estimate several quantities from a single sample, and the best importance sampling density depends on how the errors on these quantities are combined. We characterise the optimal auxiliary density for a general error criterion. We show that it is the nominal density reweighted by a weighted length of the vector of estimated functions, where the weights are set by the error criterion, or by the curvature of the loss when the criterion comes from a loss on the errors. We also provide a lower bound on the best achievable error. The optimal density is explicit for smooth losses, including the usual losses on covariance matrices. We also derive allocation laws for the cells of a histogram and characterise the minimax density for the probabilities of several events. When there is no closed form, we show how to compute the optimal density numerically.
Zihao Yuan
Fri 13 Nov
stat.MEmath.STstat.TH
pdf html
- interest Causal inference and identification in econometrics 0.90
- favorites near “When do common time series estimands have nonparametric causal meaning?” 0.87
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.89
Abstract
In empirical economics, regression discontinuity design (RDD) inference is commonly based on asymptotic normality under an independence assumption, but the running variables or disturbances may exhibit unknown and complex dependence. We first rigorously identify a fundamental dilemma for this conventional practice. To address it, we propose a validity recovery procedure that redefines the roles of bias and stochastic calibration, thereby decoupling data dependence from stochastic calibration. Taking the independence-assumption-based confidence interval as our inferential anchor, our procedure restores its validity while preserving its asymptotic leading constant, even when the data are dependent. For fixed-design RDD based on local linear regression, our intervals achieve asymptotically exact nominal coverage and, compared with the anchor, exploit higher-order smoothness without enlarging the asymptotic variance constant, attaining the minimax-optimal shrinking rates up to a mild factor, such as $\sqrt{\log T})$. The procedure requires neither specific dependence frameworks (e.g., mixing or physical dependence) nor dependence-modeling techniques (e.g., long-run variance estimation). For random-design RDD, under more specific dependence assumptions, we derive the rate-explicit coverage error at each nominal level and characterize the trade-off between this rate and interval length. Moreover, modeling dependence becomes an optional enhancement that can further enable validity recovery and improve finite-sample coverage, while averaging over repeated calibrations stabilizes the randomized center and protects against undercoverage in finite samples without increasing interval length. In short, under mild dependence and moment conditions, our procedure makes data independence an implementation default rather than a validity requirement. Simulations support the theoretical results.
Christian Gourieroux, Joann Jasiak, Cheng Zhong
Wed 18 Nov
econ.EM
pdf html
- favorites near “Spectral State Space Models” 0.85
- reading online, term structure, functions, long
- interest Online learning and statistical learning theory 0.89
Abstract
We examine and illustrate the properties of the Long Short Term Memory (LSTM) prediction models applied online to stationary inputs. We discuss the effects of: a) the network architecture that defines the admissible prediction functions, b) the objective function used to measure the prediction accuracy, and c) the Stochastic Gradient Descent (SGD) algorithm. We also study the prediction optimality, the term structure of predictions, the filtered memory dynamics. We also introduce a set of diagnostic tools for memory-based prediction models that focus on the estimated dynamics of the hidden states.
Eni Musta, Valentin Patilea
Wed 18 Nov
math.STstat.TH
pdf html
- interest Adaptive data collection and anytime-valid inference 0.88
- your work near “Efficient Online Estimation of Causal Effects by Deciding What to Observe” 0.85
- reading functions, nominal, nonparametric, large class
Abstract
We consider the nonparametric estimation of the conditional hazard, cumulative hazard and survival functions of a time-to-event variable from streaming data subject to censoring, truncation, or a cure fraction. A time-to-event model is a map from the distribution of the latent variables to that of the observed ones. For a large class of models this map has an explicit inverse, and the conditional hazard is the ratio of the density of the exactly observed durations to an at-risk function. Both are estimated by one-pass learners, in which each observation contributes once and is never revised, such as recursive kernel, series and histogram estimators. Our main result is an asymptotic confidence band sequence for the conditional hazard at a fixed covariate value, uniform in time and in the sample size, which allows monitoring after a burn-in, and stopping at a data-dependent time. The band can shrink at the optimal uniform rate up to a logarithmic factor. Bands for the cumulative hazard and the survival function follow. In simulations, a fixed-sample band inspected repeatedly fails in about half of the streams, while the miscoverage of the recommended bands stays below, or close to, the nominal level.
Jian Kang, Arnav Mazumder, Giovanni Carter Lembo, Armeen Taeb
Mon 16 Nov
stat.ME
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.90
- reading causal, partially, theory, good
- interest Adaptive data collection and anytime-valid inference 0.84
Abstract
Most work in causal discovery focuses on point estimation of the underlying causal graph, while uncertainty in the learned structure has received substantially less attention. We develop confidence regions for the completed partially directed acyclic graph (CPDAG) identified from observational data. Our approach builds on perturbed PC (Chang et al., 2026), which generates plausible CPDAGs by perturbing the conditional independence tests underlying the widely used PC algorithm. The threshold used in these tests controls how perturbations translate into variation across the resulting CPDAGs. We propose a new thresholding strategy that suppresses spurious variation across runs while preserving asymptotic coverage, producing a substantially tighter collection of plausible CPDAGs. This collection is summarized through confidence balls and model confidence intervals constructed from the partially ordered set of CPDAGs under model inclusion. We also develop new theory and algorithms for the exact computation of these regions. Simulations show that our confidence regions attain good coverage while being substantially smaller and more interpretable than the original perturbed PC collections.
Xiaoyi Zheng, Hong Zhang
Wed 18 Nov
stat.ME
pdf html
- interest Adaptive data collection and anytime-valid inference 0.91
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.86
- favorites near “Majorizing Measures, Sequential Complexities, and Online Learning” 0.87
Abstract
We study sequential conditional paired-sample testing, where observations arrive over time and the goal is to continuously assess whether two conditional distributions are equal given a common set of covariates. This problem arises naturally in model monitoring, where distributional changes must be detected without a prespecified terminal sample size, and in online comparisons, where responses are compared conditional on subject-level characteristics. We propose SCPT, an anytime-valid testing procedure for conditional distributions based on sequentially constructed conditional kernel discrepancy statistics and test martingales. The proposed method is nonparametric and accommodates a broad class of data-generating processes beyond the i.i.d. setting. We establish time-uniform Type I error control and derive asymptotic power guarantees under suitable regularity conditions. We also develop several useful extensions to address different needs, including testing under broader dependence structures, comparisons involving multiple auxiliary streams, and conditional goodness-of-fit testing. Simulations and real-data experiments show that SCPT effectively detects conditional distributional differences in streaming data, demonstrating its practical utility in a wide range of real-world applications.
Seiya Takano, Kazuki Nishida, Ryo Emoto, Shigeyuki Matsui
Mon 16 Nov
stat.ME
pdf html
- reading root, log, optimism, nominal
- interest Semiparametric inference and debiased machine learning 0.87
- favorites near “Simplifying debiased inference via automatic differentiation and probabilistic programming” 0.87
Abstract
Confidence intervals after bootstrap optimism correction are often reported for prediction models, but what they cover depends on the estimand. We distinguish two estimands: the target-population performance of the final model developed from the observed sample, $\theta_{\mathrm{FM},n}$, and the average performance over repeated model development, $\theta_{\mathrm{proc},n}$. Using first-order expansions of the estimation errors, we identify the bootstrap root that matches the error for $\theta_{\mathrm{FM},n}$. We use this root in the nested error bootstrap (NEB) as a new reference method, and we show that the optimism error bootstrap (OEB), which needs no inner bootstrap, approximates NEB. We also show that the conventional two-stage bootstrap (TSB) and location-shifted bootstrap (LSB) reproduce the first-order variation for $\theta_{\mathrm{proc},n}$. In simulations, NEB generally achieved coverage closer to the nominal level than TSB and LSB for the area under the receiver operating characteristic curve, Brier score, and log loss. OEB was close to NEB at moderate to high events per variable (EPV) but could give narrower intervals at low EPV. The distinction is particularly clear for calibration measures: first-order effects of perturbing the training and evaluation distributions cancel in the TSB root, yielding a degenerate first-order limit. In contrast, the NEB root has a nondegenerate first-order limit and yields asymptotically valid confidence intervals for $\theta_{\mathrm{FM},n}$ under regularity conditions. Consistent with theory, TSB showed substantial undercoverage for calibration even at larger sample sizes. Our findings emphasize the need to construct confidence intervals for a specified estimand and performance measure.
Stanisław M. S. Halkiewicz
Mon 16 Nov
econ.EM
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.85
- interest Causal inference and identification in econometrics 0.84
- favorites near “Difference-in-Differences with multiple time periods” 0.84
Abstract
Conventional two-way fixed-effects (TWFE) inference need not be uniformly adequate even when all treatment weights are positive. Under a local treatment-effect drift, the statistic for the equally weighted treated-cell effect has worst-case mean shift $(c/q)\Gamma$, where $\Gamma=\sqrt{N_1}\|w-u\|_2$ is design-only and $q$ is the cluster-score noise scale. Projection matches the exposure to cohort, event-time, additive, and saturated group-time pilots; when residualized treatment is constant within treated group-time cells, $\Gamma_{\mathrm{gt}}=\Gamma$. A direct score estimator is feasible with fixed panel length as clusters grow. Because a debiased quadratic pilot is nonregular at zero, we use two regular one-sided procedures: a reverse-triangle multiplier lower bound to detect unsafe envelopes and covariance-aware noncentral-$\chi^2$ inversion to certify adequacy. At $5\%$ nominal size with a five-point tolerance, the shift threshold is $0.652$. In the Callaway--Sant'Anna minimum-wage panel, where every treated-cell weight is positive, the saturated $95\%$ lower shift bound is $0.793$, so conventional inference is not uniformly certified. A complementary Brazilian property-tax reanalysis with $2{,}840$ municipalities withholds certification in all four classes; its smallest lower bound is $1.040$. Under the maintained many-cluster approximation, castle-doctrine homicide passes for additive cohort-plus-event-time heterogeneity with a $95\%$ upper bound of $0.236$; its saturated covariance lacks one required rank. Divorce is inconclusive. These envelopes are not predictions of realized rejection. The diagnostic separates point correction, uniform exposure, and statistically warranted certification.
Sergey Porotsky
Fri 13 Nov
stat.ME
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.93
- interest Causal inference and identification in econometrics 0.85
- reading observational, observational data, presents, causal
Abstract
Unmeasured confounding factors in observational data may lead to biased estimates of the effect of an exposure on an outcome. Consequently, a central problem in causal inference based on observational data is sensitivity analysis with respect to unmeasured confounding. Existing sensitivity-analysis methods generally focus on worst-case bounds. We propose an exact method for quantifying the significance of confounding, defined here in terms of the complete range of analytical values of the stratification-based Risk Ratio over the set of all joint distributions compatible with the observed characteristics. We refer to the proposed method as Inverse Confounding Analysis (ICA) - it is described in [1]. This paper presents examples of applying ICA to several real-world statistical datasets and examines an important issue that has attracted criticism of the method: the practical feasibility of obtaining the input data required for its application. The results are compared with those of two other methods for confounding analysis, the E-value and the method proposed by Pena.
Tung Dang, The Hung Phung, Son Lam Nguyen, Tu Nguyen
Wed 18 Nov
cs.LGcs.CEecon.EMstat.AP
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.86
- favorites near “Dynamically Optimal Treatment Allocation” 0.83
- reading policy, monetary, log, nominal
Abstract
Prescriptive analytics initiatives in logistics, such as offline reinforcement learning and next-best-action systems, are often funded before anyone verifies that the available historical logs can support an improved decision policy. We formalise this pre-investment feasibility question via action-outcome coupling: a total-effect variance index measuring how much expected outcomes change with the logged action. Coupling provides an algorithm-independent upper bound on any policy's achievable value, expressed in monetary units per decision for both single-step and discounted sequential objectives. We also analyse the asymptotic false-licensing risk of an invalid workflow: fitting a value model and evaluating the model's greedy policy on the same data. The false-licence rate is driven mainly by the number of state cells over which selection is performed, rising from the nominal test level with one cell to near certainty with a few dozen, while overlap has a smaller effect. We then propose an audit that assigns logs to four tiers of decision support, separately for fixed and contextual policies, and returns an explicit stopping rationale. On a randomised e-commerce log, the audit endorses a fixed recommendation policy, certifying gains equal to 30 per cent of the theoretical ceiling; the remaining gap reflects bound looseness under an 80-action design. Two public logistics logs fail at different checkpoints: one because the target is an accounting identity of the action, and another because, with 27 clusters, estimated coupling stays below the design's detectability threshold even though a conservative offline learner certifies an improvement in 29 of 30 seeds.
Víctor Gallego
Tue 17 Nov
stat.COcs.LGstat.ML
pdf html
- interest Computational methods for heterogeneous agent and dynamic economic models 0.82
- favorites near “Monte Carlo Geometry Processing” 0.83
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.82
Abstract
In structured Bayesian models, a large latent block is often Gaussian conditional on a smaller set of parameters, yet generic inference methods relearn this geometry during warmup. We study how this structure can be used to construct transports for posterior inference. The conditional mean and precision determine exact affine maps, including symmetric Brenier and triangular Knothe-Rosenblatt maps, that separate a standard normal block from the remaining non-Gaussian core. KL and relative Fisher objectives then reduce to objectives on this core. We introduce CONGA, a compiler that identifies candidate Gaussian blocks and selects a temporal, spectral, or sparse implementation according to their structure and computational cost. Marginalization requires an identity established from the model; numerical evidence alone permits only a target-preserving change of variables, and a failed check leaves the model unchanged. On the Stock-Watson inflation model, CONGA improves minimum ESS per second by 12.5x for NUTS. It also restores convergence where the raw target fails: for WALNUTS on the same model, and for NUTS on a 6760-dimensional electricity load model reduced to 34 core coordinates. Experiments on graph-indexed fields show how coloring and fill affect the choice of map, including cases where retaining the original target is preferable. Code at https://github.com/vicgalle/conga
Hao Qin, Ayumi Mizuno, Kyle Morrison, Allison K. White, Shinichi Nakagawa
Fri 13 Nov
stat.ME
pdf html
- interest Causal inference and identification in econometrics 0.91
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.90
- favorites near “Difference-in-Differences with multiple time periods” 0.87
Abstract
Generalized linear mixed models (GLMMs) are widely used across scientific disciplines, yet quantifying response variation explained by auxiliary information remains challenging for non-Gaussian outcomes and hierarchical dependence. Existing $R^2$ measures provide useful summaries, but their variance-based constructions are not generally linked to a common geometric interpretation, a fixed-design marginal estimand, and corresponding statistical inference. We propose marginal quasi-$R^2$ through two distinct operations. First, marginalization integrates over latent random effects, retaining their uncertainty in the marginal response moments and specifying the model for explanatory assessment. Second, a Quasi-GLM formulation supplies a local working information metric under which orthogonal projection yields an Pythagorean decomposition of explained and unexplained variation. We define a fixed-design population estimand, establish consistency and asymptotic normality of its estimator, and derive cluster-robust sandwich standard errors and confidence intervals under stated regularity conditions. Simulations assess estimation accuracy and interval coverage across response families and random-effects structures. Two ecological applications illustrate the importance of reporting estimation uncertainty alongside marginal quasi-$R^2$. The method is implemented in the R package quasiR2.
Mingyuan Xu, Rundong Huang, Doudou Zhou, Tongyu Li
Fri 13 Nov
stat.ME
pdf html
- interest Macroeconometrics and time series 0.89
- favorites near “Advances in Nowcasting Economic Activity: Secular Trends, Large Shocks and New Data” 0.82
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.79
Abstract
We develop inference for structural patterns in the temporal association of object-valued time series. Our framework is built on pairwise lag profiles, which record the expected value of a chosen pairwise function at each lag and serve as a unified inferential target. We consider three representative structural features: whether a profile reaches stability beyond a finite lag, remains constant over time, and displays geometric regularity at short lags. To assess these properties, we design statistical tests calibrated by a common block multiplier bootstrap that accommodates serial dependence and the overlap inherent in lagged pairs. Gaussian approximation results establish bootstrap validity for asymptotic size control, while suitable signal conditions guarantee consistency against the corresponding alternatives. Finite-sample performance is examined through simulations in both Euclidean and density-valued settings. Applications to electricity load curves and stock-return distributions illustrate how the proposed procedures characterize association across lags.
Zhen Fang, Guanting Chen, Yufeng Liu
Fri 13 Nov
stat.MLcs.LG
pdf html
- your work near “Efficient Online Estimation of Causal Effects by Deciding What to Observe” 0.87
- interest Reinforcement learning, policy learning and off-policy evaluation 0.84
- reading treatment, theory, regret, randomized
Abstract
Learning a risk-sensitive individualized treatment rule (ITR) with patient covariates is challenging when the treatment criterion depends on the full conditional outcome distribution rather than only its conditional mean. Existing methods are largely criterion-specific, making it difficult to compare multiple treatment criteria within a common pipeline, such as quantiles, conditional value-at-risk (CVaR), and complex preference-based objectives such as cumulative prospect theory. We propose a distribution-first framework for risk-sensitive ITR learning that first estimates treatment-specific conditional outcome distributions and then constructs the target ITR by plug-in criterion evaluation. The proposed framework unifies treatment rule learning across multiple criteria and accommodates hard-to-optimize objectives within the same estimation pipeline. On the theoretical side, it yields a general error propagation result for locally H\"older-stable criteria, linking first-stage distribution estimation error to downstream criterion estimation and treatment regret, together with sharper regret bounds under standard margin conditions. Empirically, we show that the proposed framework outperforms direct criterion-specific methods in multiple settings and further evaluate it on the ACTG175 HIV clinical trial, a randomized study comparing several antiretroviral regimens in HIV-infected adults.
Luca Scaffidi Domianello
Fri 13 Nov
stat.CO
pdf html
- favorites near “Simplifying debiased inference via automatic differentiation and probabilistic programming” 0.92
- interest Semiparametric inference and debiased machine learning 0.91
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.82
Abstract
Maximum Likelihood Estimates (MLEs) of the parameters of a Multiplicative Error Model are typically computed through a standard gradient method, an iterative procedure usually based on numerical approximations to derivatives of the log-likelihood function, owing to the complexity of the analytical derivatives, which involve recursive terms. Building on previous work developed mainly in the GARCH framework, we derive analytical expressions for the score vector and the Hessian matrix of the Multiplicative Error Model log-likelihood by adopting the Gamma distribution for estimation purposes. Then, we illustrate how the iterative algorithm based on the analytical derivatives should be computationally implemented. After maximizing the log-likelihood, we also provide explicit formulas for several estimators of the asymptotic covariance matrix of the MLE. Additionally, we provide analytic formulas for robust covariance estimators that remain valid under distributional misspecification, which are also referred to as the asymptotic variance of the Quasi-Maximum Likelihood Estimator. We also propose a covariance estimator that is consistent under general forms of heteroskedasticity of the error term, while the usually adopted estimators rely on a specific variance assumption. The finite-sample performance of the covariance estimators is assessed through a Monte Carlo investigation. An application on real data completes the analysis.
Sujit K. Ghosh, Timothy Boakye, Kimberly F. Sellers
Mon 16 Nov
stat.ME
pdf html
- reading semiparametric, moment, adjustment, empirical
- your work near “Valid Inference with Imperfect Synthetic Data” 0.81
- interest Bayesian computation and structural estimation 0.85
Abstract
We develop a Bayesian semiparametric regression framework for count data that can accommodate both underdispersion and overdispersion without specifying a full parametric probability model. The model is built from the first two conditional factorial moments of a zero altered distribution. Covariates may enter the conditional mean and the conditional dispersion separately, thereby allowing the dispersion to vary across observations. We turn the moment restrictions into an adjusted exponentially tilted empirical likelihood (AETEL) criterion and combine this criterion with Gaussian priors to form a Bayesian empirical-likelihood pseudo-posterior. The adjustment improves numerical feasibility and a two-block Metropolis delayed acceptance sampler allows for faster runs of the MCMC chains. This reduces unnecessary exact evaluations without changing the AETEL target. Using existing Bayesian exponentially tilted empirical likelihood (ETEL) theory, we state a first-order posterior normal approximation and verify the moment, smoothness, and Jacobian rank conditions needed for the asymptotic inference. Simulations under zero inflated Poisson and Conway Maxwell Poisson data exhibit small bias, accurate uncertainty estimates, and near nominal interval coverage for both overdispersed and underdispersed counts. Applications to various datasets illustrate how the Bayesian analysis can identify covariate dependent dispersion using finite sample posterior inference.
Diego Silva Prado, Bruno Caldas de Souza, Tiago Amaral Penteado de Jesus, Ricardo Guimarães Herrmann, Cristiano Breuel, Cristiano Bevitori Maffia De Oliveira
Mon 16 Nov
cs.LGq-fin.RMstat.ML
pdf html
- interest Reinforcement learning, policy learning and off-policy evaluation 0.87
- favorites near “Finding Regularized Competitive Equilibria of Heterogeneous Agent Macroeconomic Models via Reinforcement Learning” 0.81
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.82
Abstract
When a sequential decision system falls short of an optimum, its score does not reveal which modifications would improve its decisions. DECODE is a benchmark for testing such modifications through controlled experiments on planning and available information. Its consumer-credit collections generator provides five partially observed environments that share known dynamics matrices and vary a memory parameter and process noise. The experiments change selected inputs or planning rules while keeping the rest of the solver fixed. Each comparison specifies the solver's state inputs, response inputs and planning horizon. On the deterministic base environment, which was selected for a large gap between six-month and one-month planning with the hidden state, extending planning from one to six months increases the scores of two gradient-boosted solvers by 15 to 17 percentage points of the exact optimum, computed with the hidden state and borrower traits. Even at this horizon, the best graded solver, whose repayment response is fitted on logs, remains 15.6 points below that optimum. In a specified intervention sequence, three changes to its state inputs and their evaluation during search lower its score by 6.3 points, and replacing its fitted repayment model with the true response, evaluated with hidden borrower traits, then raises it by 15.6, a net recovery of nine of those points. Because the intermediate effects are conditional on the repayment model and the order of the changes, they do not uniquely assign the gap to separate causes. Within these limits, DECODE makes hypotheses about solver performance testable through specified interventions under known dynamics and explicit information conditions.
Christa Cuchiero, Francesca Primavera, Sara Svaluto-Ferro
Fri 13 Nov
math.PRstat.ML
pdf html
- reading odes, finite-dimensional, procedures, series
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.84
- interest Computational methods for heterogeneous agent and dynamic economic models 0.79
Abstract
We develop score-based generative models for time series and standard finite-dimensional data, building on the analytical computation of the expected signature of a new class of jump-diffusions, called entire sig-jump-diffusions. These are semimartingales whose characteristics are given by linear combinations of potentially infinitely many components of their own signature. We show that their expected signature can be computed by solving linear ODEs in the extended tensor algebra, providing a tractable foundation for learning procedures. Specifically, the learning of a generalized score function for jump diffusions, a key ingredient of our generative modeling framework, can be reduced to solving linear ODEs and a deterministic optimization problem. We further prove that polynomial jump-diffusions can be embedded into this framework, and that their expected signature reduces to a finite-dimensional system of linear ODEs. Building on this result, we develop an alternative approach for finite-dimensional data, based on inverse signature matching, a method that learns the score by matching the expected signature of the reverse process with that of the inverse forward signature.
Alessandro Baldi Antognini, Rosamarie Frieri, Maroussa Zagoraiou
Tue 17 Nov
stat.ME
pdf html
- reading nonparametric, semiparametric, functional, treatment
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.87
- interest Reinforcement learning, policy learning and off-policy evaluation 0.83
Abstract
Response-Adaptive Randomization (RAR) assigns incoming patients to treatments with probabilities that depend on previously observed responses. While most existing RAR procedures rely on finite-dimensional summaries of treatment performance (such as rates, means, or model parameters), we develop a nonparametric (NP) distribution-based framework in which the treatment outcome distributions are learned sequentially and adaptation is driven by their functionals. The framework naturally accommodates multiple treatments and is particularly suited to endpoints for which average effects may conceal relevant differences across the outcome distribution. We introduce an ethical NP-RAR rule driven by a pre-specified distributional score and a nonparametric extension of the multi-arm efficient randomized adaptive design (NP-ERADE) able to approach generic distribution-dependent targets selected according to ethical or inferential optimality criteria. We establish strong consistency and joint functional central limit theorems for the empirical treatment distributions, together with asymptotic normality of functional scores, estimated targets and allocation proportions. We further show that the NP-ERADE allocation proportions attain the semiparametric efficiency bound for the target. We develop post-trial inference for score contrasts and distributional features, including crossing locations, and investigate finite-sample properties, robustness and inferential performance via simulations.
Andrew C. Parnell
Mon 16 Nov
stat.APstat.COstat.ME
pdf html
- reading economic, income, tax, bayesian causal
- favorites near “When Did Growth Begin? New Estimates of Productivity Growth in England from 1250 to 1870” 0.80
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.80
Abstract
When a government increases public spending on tertiary (university) education, does it provide a positive fiscal return on investment once general economic growth is separated out? Cross-country data make this hard to answer because growing economies naturally collect more tax, international ratios divide per-student spending by GDP per person and student headcount, and cohorts take years to graduate. We assemble a five-year panel of 121 countries (1985-2024) and fit continuous-treatment Multivariate Bayesian Causal Forests that model four tax changes jointly whilst adjusting for starting conditions, predicted funding, and follow-up economic growth. A +10 percentage-point (pp) increase in per-student university spending (% of GDP per person) shifts the tax mix toward direct income and profit taxation by +0.36 pp after 5 years (93% probability of a positive effect) and +0.71 pp after 10 years (96% probability). Overall tax-to-GDP takes a decade to respond as cohorts graduate: 10-year total non-resource tax rises by +0.24 pp of GDP (89% probability) and direct income tax by +0.12 pp (90% probability) per +10 pp of per-student spending, by +0.37 pp (100% probability) after five-year within-country funding increases, and by +0.22 pp in total tax (95% probability) and +0.15 pp in direct income tax (98% probability) per +25% increase in real PPP dollars per student. Comparing one entering cohort's degree cost against its 40-year career tax payments shows a typical three- to five-fold total tax return in high-income economies (EUR 2.37 billion in Ireland returning EUR 9.66 billion in total tax with a 77% probability of net profit and EUR 2.87 billion in direct income tax, or EUR 11.55 billion in the UK returning EUR 58.33 billion), whereas expanding the university budget without protecting spending per student yields no return.
Anthony Frazier, Siyu Heng, Wen Zhou
Thu 19 Nov
stat.ME
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.93
- interest Semiparametric inference and debiased machine learning 0.89
- favorites near “Difference-in-Differences with multiple time periods” 0.83
Abstract
In observational studies, unmeasured confounding may bias causal conclusions even after matching on measured confounders. Sensitivity analysis is therefore usually needed to assess the robustness of causal conclusions to unmeasured confounding. We consider multivariate sensitivity analysis that assesses the causal effect of a single treatment on multiple outcomes. We identify an inherent feature of such analyses, which we call ``confounding heterogeneity'': different outcomes may be affected by distinct unmeasured confounders, and a common unmeasured confounder may affect different outcomes to varying degrees. Building on matched observational studies, we formulate sensitivity analysis frameworks that accommodate confounding heterogeneity. We show that Type I error control for many existing sensitivity analysis methods is guaranteed under sensitivity analysis models and modes of inference that differ fundamentally from those used in the classic framework. We study asymptotic power using design sensitivity and Bahadur-Rosenbaum efficiency and examine finite-sample power through numerical studies. Our results show that the structure of unmeasured confounding affects sensitivity analysis, and that widely used single-parameter frameworks may mischaracterize the sensitivity of a causal conclusion to unmeasured confounding when confounding heterogeneity is considered. Simulations and a data application illustrate the practical importance of confounding heterogeneity in multivariate sensitivity analyses.
Hao Fu, Xinbing Kong, Zeyu Li, Junfan Mao
Tue 17 Nov
stat.MLcs.LG
pdf html
- favorites near “Eight Centuries of Global Real Interest Rates, R-G, and the 'Suprasecular' Decline, 1311-2018” 0.79
- reading asset, asset pricing, firms, returns
- interest Macroeconometrics and time series 0.83
Abstract
In this paper, we propose a conditional asset pricing transformer (CAP-T) that captures nonlinear interactions among firms through attention to forecast excess returns. In the first stage, a transformer estimates conditional expected returns, and kernel-weighted principal component analysis of its residuals estimates the local factor loading space. Projecting the conditional mean onto this space separates factor risk compensation from conditional mispricing. Then, in the second stage, separate transformers learn these two components, whose predictions are combined to forecast returns. As a result, the proposed framework decomposes returns into conditional mispricing, factor risk compensation, a centered factor component, and an unpredictable innovation. We establish approximation and generalization bounds and characterize how first-stage prediction errors propagate through local factor estimation and component learning, providing statistical guaranties for conditional mispricing, factor loadings, and factor risk premia. Finally, simulations and empirical studies demonstrate the predictive and interpretive benefits of the proposed method.
Hongqiang Sun, Sayar Karmakar, George Michailidis
Fri 13 Nov
stat.MEecon.EM
pdf html
- favorites near “Priors for the Long Run” 0.90
- interest Macroeconometrics and time series 0.88
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.79
Abstract
We study the localization of mean changes in high-dimensional vector autoregressions (VARs) when the autoregressive dynamics remain common across regimes. We develop a three-step procedure that first obtains preliminary change-point estimates from the raw series ignoring the dependence structure, then estimates a common sparse VAR using observations away from the estimated changes, and finally filters the series to refine the change point locations. For a single change point, we show that the preliminary localization rate depends directly on the strength of temporal dependence, whereas filtering removes this dependence penalty from the oracle rate. With an estimated filter, the refined rate additionally reflects the error in estimating the VAR dynamics and the relation between the change in the stationary mean and the corresponding change in the VAR intercept. We characterize these rates explicitly and establish conditions under which filtering improves localization. Because filtering near a change point uses lags from adjacent regimes, it also creates a short-lived distortion in the filtered mean; we characterize when this boundary effect is innocuous and develop a transition-corrected alternative. We extend the procedure to an unknown number of change points. Overlapping local windows are used to generate candidate locations, an information criterion removes redundant and spurious candidates and estimates the number of changes, and the retained locations are subsequently refined using the estimated VAR filter. We establish consistency for the estimated number of changes and simultaneous localization rates for their locations. Numerical experiments corroborate the localization gains and show robustness to moderate departures from common VAR dynamics. An application to a large panel of U.S. macroeconomic indicators identifies economically interpretable changes.
Athanasios Christou Micheas
Tue 17 Nov
math.PRmath.FAmath.STstat.TH
pdf html
- reading functions, processes, taylor, motion
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.81
- favorites near “Introduction to regularity structures” 0.85
Abstract
We introduce stochastic multivariate Taylor measure functions (SMTMFs), which extend the multivariate Taylor measure function framework through random Taylor coefficients and a possibly random Taylor parameter. We establish their measurability properties and derive sufficient conditions for the associated processes to be martingales, including criteria based on conditional covariance and continuous semimartingale decompositions. We construct standard Brownian motion as an almost surely uniformly convergent limit of finite SMTMFs using the Levy-Ciesielski representation. We also show that the completion of the finite stochastic MTMF space under the expected uniform norm contains Brownian motion, Brownian bridges, fractional Brownian motion with Hurst parameter H, and Ornstein-Uhlenbeck processes with integrable initial conditions. Our main result characterizes this completion as the Bochner space, under an admissibility condition on jointly measurable stochastic Taylor coefficients. The proof uses dyadic piecewise linear interpolation to approximate integrable continuous stochastic processes in the expected uniform norm. Thus, the completed SMTMF space encompasses the full class of integrable continuous random functions on a compact time interval.
Ryan Lim
Tue 17 Nov
stat.MEcs.LGmath.STstat.MLstat.TH
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.86
- interest Semiparametric inference and debiased machine learning 0.86
- reading units, rational, large, control
Abstract
Weighted conformal p-values generalize conformal inference to cases with covariate shift. One way to use them is to pass them to the Benjamini-Hochberg (BH) procedure, either to decide which test units to select or to detect outliers. But weighted conformal p-values may not be positively dependent, so we want to know: does BH still control the false discovery rate (FDR) in finite samples? Based on experiments, it appears to, but we prove it doesn't. We give two counterexamples, certified in exact rational arithmetic, where the FDR exceeds its target level $q$. We also identify how large the difference between the FDR and $q$ can be. With $n$ calibration points, $m$ test units and weight ratio $B$, the relative excess $\mathrm{FDR}/q-1$ is bounded by a function of $n$, $m$, $B$ and $q$ that is of order $\sqrt{Bm/(nq)}$ when $Bm/(nq)$ is small. The bound holds for every configuration with $B\le n$ and every $n$. This order can be reached when multiple non-null test units share a common threshold: in a Poisson-limit model, that configuration's excess is $c(q)\sqrt{Bm/(nq)}$ to first order as $B/n\to0$ and then $m\to\infty$. For some fixed distributions with this structure, the excess decreases only like $n^{-1/2}$ as $n$ grows. If every test unit is null, the excess is $O(B/n)$ at fixed $m$, and in certain configurations it grows linearly with $m$. To assist practitioners in judging how large the FDR can be, we turn that bound, which depends only on $n$, $m$, $q$ and $B$, into a lowered BH level with proved finite-sample control, and give a much less conservative rule of thumb that covers every worst case we were able to find but provides no guarantee.
Subhodeep Dey, Gopal K. Basak, Samarjit Das
Tue 17 Nov
stat.ME
pdf html
- favorites near “Measuring Price Selection in Microdata: It's Not There” 0.87
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.85
- interest Causal inference and identification in econometrics 0.86
Abstract
In clustered data, cross-sectional dependence within clusters can be characterized as weak, semi-strong, or strong. This paper proposes a simple test to identify the specific nature of this dependence in linear regression models. By exploiting the variation in cluster sizes, our approach detects the underlying dependence structure in a computationally simple and robust manner. We develop a test statistic that is shown to be asymptotically chi-square distributed under the null hypothesis of weak dependence. Simulation studies and an empirical illustration using India's Household Consumption Expenditure Survey (HCES) data for 2022-23 demonstrate the usefulness of the proposed test.
Chung Eun Lee, Xianyang Zhang
Fri 13 Nov
stat.ME
pdf html
- interest Semiparametric inference and debiased machine learning 0.90
- your work near “Valid Inference with Imperfect Synthetic Data” 0.84
- favorites near “Priors for the Long Run” 0.84
Abstract
This article introduces a novel nonparametric test for conditional mean independence, which formally evaluates whether a random vector X has an additional impact on the mean regression function of a response Y when a conditioning random vector Z is already included. By framing the problem as classical partial effect testing, our approach goes beyond traditional parametric models to provide a highly general and flexible framework. The test statistic is constructed using martingale difference divergence (MDD) after neutralizing the effect of Z on Y via a sieve approximation. The population residual MDD characterizes conditional mean independence under suitable moment conditions. We present a theoretical framework for the test under both a fixed number of basis functions and a growing sieve dimension. We derive the nonstandard and non-pivotal asymptotic null distributions in both regimes. To facilitate practical implementation, we propose a wild bootstrap procedure to approximate these limiting distributions and establish its asymptotic consistency. To establish the fundamental properties required for the test to be widely useful, we conduct a power analysis, demonstrating the test's consistency under fixed alternatives and establishing its asymptotic power against local alternatives. The practical utility and finite-sample performance of the proposed test are examined through simulation studies and an exploratory application to a real-world variable selection problem in regression modeling. The code for the proposed test and numerical simulations can be found at: https://github.com/leece087/Conditional-Mean-Independence-Test.
Gianluca Cubadda, Luca Margaritella, Orest Prifti
Mon 16 Nov
econ.EMstat.ME
pdf html
- favorites near “Priors for the Long Run” 0.89
- interest Macroeconometrics and time series 0.88
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.78
Abstract
We study regularized estimation in high-dimensional vector autoregressions under physical dependence and row-wise approximate sparsity using the Broken Adaptive Ridge (BAR) estimator. BAR is an iteratively reweighted Ridge procedure whose fixed-point first-order conditions imply a coordinatewise thresholding rule that produces exact zeros. We show that the BAR estimation error can be analyzed on the union of the population support above the BAR threshold and the support selected in the sample. The size of this union can be controlled at the rate implied by approximate sparsity, allowing us to establish the required restricted-curvature condition and derive L1, L2, and prediction error rates when the number of regressors grows polynomially with the sample size and may exceed it. Under exact sparsity, the resulting rates recover the familiar rates for sparse high-dimensional estimation. Monte Carlo experiments across a range of high-dimensional VAR designs show that BAR delivers strong estimation and forecasting performance, and retains a high share of relevant coefficients relative to Lasso, adaptive Lasso, and SCAD. An application to U.S. inflation forecasting using FRED-MD shows particularly strong performance at the one-month horizon.
Yannick Baraud
Fri 13 Nov
math.STstat.TH
pdf html
- interest Semiparametric inference and debiased machine learning 0.90
- reading functions, moment, absolute, parametric rate
- your work near “Computational Methods for Economic Models with Function Valued States” 0.85
Abstract
We compare the least squares estimator (LSE) with the $\ell_{1}$-estimator recently introduced by Baraud and Maillard~\citeyearpar{BM-2026} for estimating a regression function under a fixed design, when the errors are independent but possibly heteroscedastic and heavy-tailed. We show that the two estimators coincide on the class of nondecreasing (or nonincreasing) functions on an interval. As a consequence, the isotonic LSE inherits the non-asymptotic $\ell_{1}$-risk bounds of the $\ell_{1}$-estimator, which only require a moment of order $p\in[1,2]$ on the errors and depend on the average of their $p$-th absolute moments rather than on their maximum. This coincidence no longer holds on the nonconvex class of unimodal functions: we exhibit a heteroscedastic Gaussian example in which the LSE is inconsistent while the $\ell_{1}$-estimator converges at the parametric rate. Finally, we show on a one-dimensional linear model that neither estimator dominates the other: depending on the way the noise is distributed among the design points, the LSE may converge at rate $1/n$ while the $\ell_{1}$-estimator is inconsistent.
Yunda Hao, Kin Wai Chan
Thu 19 Nov
stat.MEmath.STstat.TH
pdf html
- favorites near “Orthogonal Statistical Learning” 0.90
- interest Adaptive data collection and anytime-valid inference 0.86
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.82
Abstract
To perform an anytime-valid goodness-of-fit test, we propose spacing-based \(e\)-processes that control the type-I error rate under arbitrary stopping rules. Based on the maximum spacing of order statistics, we construct spacing-based \(e\)-processes for nonparametrically testing distribution. The resulting procedures are simple and computationally efficient, and are power-consistent. We call them $e$-spacing tests. This approach is especially powerful in detecting localized discrepancy between the data-generating density and the null density. Under mild conditions, the logarithm of the proposed \(e\)-process is asymptotically equivalent to that of the oracle-optimal \(e\)-process. We further generalize the construction from maximal spacing to maximum spacing set in the form of the \(1/k\) fraction of the largest \(m\)th order spacings, as a measure of the lower-level set of density. It boosts power under more global discrepancy from the null. Furthermore, we propose a greedy predictable scheme, via argmax or softmax rules, to combine multiple $e$-spacing testing strategies. We prove that it optimally tunes the $e$-spacing test. The proposed approach is readily generalizable to multivariate testing via a coordinate-wise Rosenblatt-transform. Simulation studies demonstrate that the proposed methods perform favorably relative to the considered anytime-valid competitors as well as various types of multiple test-adjusted Kolmogorov--Smirnov tests.
Duy Nguyen
Wed 18 Nov
math.STecon.EMmath.PRstat.TH
pdf html
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.85
- favorites near “Priors for the Long Run” 0.84
- interest Bayesian computation and structural estimation 0.86
Abstract
We establish the asymptotic theory of the exact maximum likelihood estimator (MLE) for discretely observed multivariate ergodic diffusions under high-frequency, long-span sampling, solving an open problem posed by \citet{YCW2019} and \citet{Li2013}. The drift may grow linearly, and the volatility is bounded away from zero and either bounded or proportional to $1+|x|$. When the sampling interval decays polynomially, the exact MLE is consistent, asymptotically normal and efficient, with rates $\sqrt{n\Delta_n}$ for drift and $\sqrt n$ for diffusion parameters. Under an additional condition on ergodic averages, and when the time span $n\Delta_n$ also grows polynomially, its moments converge as well. The proof never analyzes the exact likelihood directly: it shows that a closed-form density expansion of sufficiently high order is uniformly accurate in relative terms at every observed transition, so that the two log-likelihoods differ by $o_P(1)$ uniformly in the parameter. The same argument yields accuracy conditions under which an approximate likelihood is asymptotically equivalent to the exact one; in particular, the approximate MLE of Yang, Chen and Wan converges to the true parameter. For the delta-expansion estimator of fixed order $K\ge1$, we derive the leading bias explicitly. It shows that $n\Delta_n^{K+1}\to0$ cannot be weakened when the bias constant is nonzero, and that the bias falls on the drift parameters for even $K$ and on the diffusion parameters for odd $K$. Monte Carlo experiments confirm the predicted biases.
Axel Bücher, Mario Krali
Mon 16 Nov
math.STstat.MEstat.TH
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.87
- interest Semiparametric inference and debiased machine learning 0.87
- favorites near “Empirical Bayes for the Reluctant Frequentist” 0.89
Abstract
The causal tail coefficient (CTC) measures the influence of an extreme value of one variable on another variable and has been proposed as a tool for causal discovery in heavy-tailed models. Despite its popularity, the asymptotic behavior of its empirical counterpart has not been well understood. We develop an asymptotic theory for the rank-based CTC estimator with an intermediate threshold under nonparametric conditions that are satisfied by all standard copula models. A key observation is that the CTC is determined by the mean of a conditional tail limit distribution, which clarifies its interpretation and determines the asymptotic variance of the estimator. We establish joint asymptotic normality for the entire CTC matrix and prove consistency of a multiplier bootstrap for simultaneous inference. For heavy-tailed linear and recursive max-linear structural equation models, we show that the CTC equals one if and only if the first variable is an ancestor of the second, which leads to confidence graphs for the ancestral graph with asymptotic coverage guarantees. All results extend to general conditional tail coefficients, defined as further functionals of the conditional tail limit distribution. A simulation study and an application to dietary intake data illustrate the methods.
Jingrun Chen, Yixin Luo, Xiaojun Peng
Tue 17 Nov
math.NAcs.NAstat.ML
pdf html
- favorites near “Neural Operator: Learning Maps Between Function Spaces With Applications to PDEs” 0.86
- reading equations, hilbert space, differential equations, pdes
- interest Kernels, operators and functional data 0.86
Abstract
The Random Feature Method (RFM) distinguishes itself among machine learning approaches to partial differential equations (PDEs) through its superior accuracy and computational efficiency. Nevertheless, the theoretical underpinnings of RFM remain incompletely elucidated. This paper provides a comprehensive analysis of RFM within the framework of kernel methods, focusing on two principal aspects: (i) approximation by finitely many random features in Sobolev norms and the PDE residual seminorm, and (ii) estimation over the induced reproducing kernel Hilbert space (RKHS) from finitely many noiseless collocation observations. We relate both problems to kernel ridge regression (KRR) and establish upper and lower bounds in terms of the eigenvalues of the associated integral operators under the respective assumptions. Combining the approximation and estimation lower bounds, we establish a lower bound under simultaneous restrictions on the numbers of random features and collocation observations. We further quantify how activation regularity affects these eigenvalues, obtaining algebraic upper bounds for finitely smooth activations and stretched-exponential upper bounds for analytic activations.
Haichen Hu, David Simchi-Levi
Tue 17 Nov
stat.MLcs.LGmath.PR
pdf html
- reading online, hilbert space, hilbert, kernels
- interest Online learning and statistical learning theory 0.89
- favorites near “Hierarchies of Relaxations for Online Prediction Problems with Evolving Constraints” 0.82
Abstract
We characterize, in the worst case over kernels, the sharp minimax width of online confidence intervals for elements of a reproducing kernel Hilbert space (RKHS). This resolves the open problem posed by Vakili et al. (2021) in its original discrete-time form. A learner observes $y_i=f(x_i)+\epsilon_i$ at adaptively chosen points, with $\|f\|_{\mathcal H_k}\le B$ and i.i.d. $R$-sub-Gaussian noise, and reports the interval $|f(x)-\mu_n(x)|\le\rho_n(\delta)\sigma_n(x)$, where $\mu_n$ and $\sigma_n$ are the posterior mean and standard deviation of a surrogate Gaussian process model with regularization parameter $\lambda$. Writing $\rho_n^*(\delta)$ for the minimax confidence width for a fixed kernel, we show that, for the sum of a linear kernel and a periodic Sobolev kernel, all sufficiently large $n$, and all $\delta\in(0,1/4]$, $\rho_n^*(\delta)\asymp B+(R/\lambda)\sqrt{\gamma_n+\log(1/\delta)}$, where $\gamma_n$ is the maximal information gain and grows polynomially in $n$. Building on Lattimore (2023), we construct an adaptive design on a fixed two-dimensional domain for which, under Gaussian noise, the normalized error at a prespecified point is $\Omega(B+(R/\lambda)\sqrt{\gamma_n})$ with constant probability. This rules out the conjectured polylogarithmic width in the worst case over kernels and online designs.
Pengfei Wang
Fri 13 Nov
math.STstat.MEstat.TH
pdf html
- interest Semiparametric inference and debiased machine learning 0.88
- your work near “Valid Inference with Imperfect Synthetic Data” 0.81
- favorites near “Orthogonal Statistical Learning” 0.88
Abstract
We study noncentral normal-reference inference for high-dimensional two-sample Behrens-Fisher problems. Within the Gaussian reference, a shifted null law is accurate if and only if covariance-weighted signal energy is negligible relative to the trace of the squared covariance. We establish the sharp uniform fourth-root error order over unrestricted spectra and a linear bound under spectral spread. A separate spectral-range criterion determines when signal magnitude alone suffices. An augmented spherical reference gives finite-sample calibration for Gaussian samples and asymptotically recovers Gaussian-reference local power under projection fourth-moment bounds and control of diagonal fluctuations, without a dimension-to-sample-size restriction. For unknown Gaussian covariance, explicit trace and hollow-Gram confidence bounds turn the linear approximation bound into a simultaneous, computable certificate over local effect ranges. Eigenspace signal energies describe directional power, while a two-point argument limits estimation of unknown local power. Simulations with 10,000 evaluation replications per configuration examine unequal and noncommuting covariances, severe sample imbalance, non-Gaussian observations and the informativeness of the certificate.
Abdullah Ahmed, Yun Yang
Tue 17 Nov
cs.LGcs.AIcs.NAmath.NAmath.STstat.MLstat.TH
pdf html
- favorites near “Neural Operator: Learning Maps Between Function Spaces With Applications to PDEs” 0.84
- reading odes, equations, differential equations, theory
- interest Computational methods for heterogeneous agent and dynamic economic models 0.80
Abstract
Transformers have demonstrated the ability to perform in-context learning, that is, to adapt to new tasks from a few examples without updating their parameters. Existing theory has largely focused on emulating gradient-based algorithms for convex statistical problems or establishing general function approximation guarantees without exploiting the underlying statistical structure. We study in-context parameter estimation and trajectory prediction for parametric ordinary differential equations (ODEs), where estimation is generally nonconvex. Under suitable identifiability and regularity conditions, we construct a transformer that estimates unknown system parameters from noisy trajectory observations through a two-stage approximate maximum likelihood procedure: a coarse search provides an initialization within a locally strongly convex region, followed by gradient descent with local linear contraction. For prediction, we compose this estimator with a neural approximation of the ODE solution map. We further show that, through parameter sharing, a single transformer can perform both parameter estimation and trajectory prediction simultaneously. For these constructions, we establish nonasymptotic guarantees and finite-pretraining oracle inequalities that separate errors due to finite pretraining data, finite context length, and finite-sized transformer approximation. Numerical experiments on linear and nonlinear ODEs further show that pretrained transformers can benefit from longer inference contexts without updating their weights.
Christian E. Galarza, Sergio Luis Mercado Londoño, Víctor H. Lachos
Mon 16 Nov
stat.ME
pdf html
- favorites near “Priors for the Long Run” 0.86
- interest Macroeconometrics and time series 0.86
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.81
Abstract
Seasonal autoregressive integrated moving-average (SARIMA) models typically target the conditional mean of a time series under a Gaussian error assumption. This approach may be inadequate when the innovations are asymmetric because the conditional mean no longer represents the most likely outcome. We propose the Modal SARIMA$(p,d,q)\times(P,D,Q)_s$ model, which links the conditional mode, rather than the conditional mean, to a multiplicative seasonal ARIMA structure. The innovations follow the Fernandez-Steel class of skewed distributions (SKD), whose location parameter is the mode, and the non-seasonal Modal ARIMA$(p,d,q)$ model is the special case without seasonal terms. We give the conditional log-likelihood for the Skew-Normal, Skewed Student-$t$, and Skewed Laplace members, establish the stationarity and invertibility conditions on the product polynomials, show that the recursive modal forecast is the joint mode of the future trajectory, and characterize the correction that yields the marginal $h$-step conditional mode. Monte Carlo experiments show that conditional maximum likelihood and its Wald inference are reliable when the seasonal factors are well separated, that the test of symmetry holds its level from about 200 observations, and that the modal estimator of the dynamic coefficients is as efficient as Gaussian least squares under symmetry and 1.2 to 2.6 times more efficient under asymmetry or heavy tails. We also show, analytically and by simulation, that nearly cancelling seasonal factors amplify the effect of the starting values, and that beyond one step the recursive forecast must be corrected to remain a forecast of the most probable value. In an application to the Canadian lynx trappings, a seasonal component for the ten-year cycle improves the non-seasonal fit for every member of the SKD family. The methods are implemented in the R package ModalForecast.
Zhongming Xie, Liping Zhu
Tue 17 Nov
math.PRstat.ME
pdf html
- interest Semiparametric inference and debiased machine learning 0.88
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.78
- favorites near “When Did Growth Begin? New Estimates of Productivity Growth in England from 1250 to 1870” 0.76
Abstract
Yule's nonsense correlation persists for Spearman's rho and Kendall's tau between independent random walks. Neither coefficient concentrates at zero as the sample size grows. For centred random walks with finite increment variances, we derive a universal joint limit expressed through Brownian occupation functionals. Integral formulas give limiting variances of approximately 0.222 and 0.108, respectively, below the Pearson value of 0.241. Rank correlations thus reduce the limiting dispersion on their respective coefficient scales. We prove that both marginal limits have symmetric densities and represent them exactly as limits of polynomial densities determined by Gaussian moment integrals. Spearman's conventional t test and Kendall's normal approximation are asymptotically invalid, with rejection probabilities tending to one under process independence.
Minor Bonilla-Gomez, Javier Trejos-Zelaya
Fri 13 Nov
stat.CO
pdf html
- reading econometric, equations, procedures, julia
- favorites near “Data Science at the Singularity” 0.85
- interest Computational methods for heterogeneous agent and dynamic economic models 0.79
Abstract
We present NEVEN, an open-source infrastructure that embeds R, Julia, Python, Quarto, and JavaScript visualization libraries into Excel through native worksheet formulas. Users run econometric estimators, solve differential equations, and compile reproducible reports using expressions such as =R.MR_Lineal(...), =J.Optimize(...), and =P.ai_call(...), without procedural programming or external application switching. The platform offers over 70 R statistical procedures, 70 high-level Julia procedures backed by more than 200 mathematical functions, and full Python 3 integration for statistical computing and numerical analysis. Each language runtime runs in an independent operating-system process connected to the Excel XLL host via Windows Named Pipes and Protocol Buffers. Fatal errors terminate only the affected child process, leaving Excel operational. A static security sandbox checks user code against 25-30 restricted patterns per language, blocking shell execution, file deletion, unauthorized network access, and dynamic evaluation through five bypass-prevention mechanisms validated by 133 tests. An embedded WebView2 engine displays interactive Plotly charts, D3.js network and hierarchical visualizations, Leaflet maps, pivot tables, LaTeX notation, PDF documents, and Pluto.jl reactive notebooks through a unified formula, =NEVEN.v(). The docked NEVEN Studio environment offers nine specialized modules, including a Chart Gallery with 325 templates, an in-memory DuckDB SQL console, a retrieval-augmented AI assistant, and a multi-language script editor with English and Spanish localization. Econometric estimates match published benchmarks to four decimal places, while Julia linear algebra decompositions achieve machine-precision accuracy (relative error below 10^-14). Released under GNU General Public License v3, NEVEN includes 342 automated unit, property-based, and integration tests.
Felix Zhou
Fri 13 Nov
math.STcs.DScs.LGstat.MLstat.TH
pdf html
- interest Semiparametric inference and debiased machine learning 0.88
- your work near “Efficient Online Estimation of Causal Effects by Deciding What to Observe” 0.84
- reading theory, believe, moment, viewed
Abstract
We revisit the problem of estimating $k$ linear regressors with self-selection bias in $n$ dimensions up to error $\varepsilon$ with the maximum self-selection criterion, as introduced by Cherapanamjeri, Daskalakis, Ilyas, and Zampetakis [CDIZ23, STOC]. Our main result is a $\mathrm{poly}(n,k,1/\varepsilon)$ time algorithm for the unknown-index variant of this problem, thereby improving upon the running times of the best known algorithms due to Gaitonde and Mossel [GM24] and Kalavasis, Mehrotra, and Zhou [KMZ26, COLT] with exponential dependence on $k$. Our algorithm succeeds in the presence of correlated noise between regressors, which captures the related setting of max-affine regression. Consequently, we also give the first polynomial-time algorithm for this problem, improving on the prior works of Ghosh, Pananjady, Guntuboyina, and Ramchandran [GPGR22, IEEE Transactions on Information Theory] and Gaitonde and Mossel [GM24] with exponential dependence on $k$. Prior works exhaustively search over a $k$-dimensional subspace spanned by the regressors, using conditional moments of the response to test candidates. In contrast, our algorithm is driven by an improved understanding of the moments of the covariates. We show that, restricting to observations with large response values, the degree-$(\log n)$ Hermite moment of the covariates approximately decomposes as a sum of tensor powers of the regressors. The first step of our algorithm can be viewed as a tensor decomposition algorithm for these implicitly represented high-order tensors, which outputs unit vectors parallel to the regressors. A secondary step based on a simple variant of ordinary least squares extracts estimates for the regressors given these unit vectors. We believe our analytic insight and algorithmic components may have applications to designing efficient algorithms for other latent-variable problems.
Ching-Chia Chou, Jeng-Min Chiou
Fri 13 Nov
stat.MEstat.AP
pdf html
- interest Causal inference and identification in econometrics 0.86
- favorites near “When do common time series estimands have nonparametric causal meaning?” 0.85
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.77
Abstract
In many applications, variability itself carries important predictive information beyond the mean, and its association with an outcome may further depend on the underlying mean trajectory. However, existing functional regression approaches primarily rely on mean-based representations of functional predictors, while variability is often ignored or reduced to scalar summaries. We propose a functional regression framework that incorporates latent mean and variance processes as predictors, together with their mean-variance interaction, allowing the predictive contribution of variability to depend on the underlying mean trajectory. We establish consistency and convergence rates for the estimated coefficient functions and interaction surface, accounting for reconstruction errors, increasing observation frequency, and diverging truncation dimensions without requiring independence between the latent processes. Applications to real data demonstrate substantial gains in predictive accuracy from incorporating the variance process, with the proposed approach outperforming competing functional regression methods in the Tecator application and revealing localized mean-variance interaction patterns.
Peida Li
Fri 13 Nov
stat.MEstat.ML
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.91
- reading structural, nuisance, reports, gives
- interest Causal inference and identification in econometrics 0.84
Abstract
Targeted covariance changes couple structural rows through a common directed acyclic graph. We study the graph support determined by these experiments when structural errors may be correlated. A positive-gap four-node family has a shortcut identified by common acyclicity. A response indegree of at most two additionally gives unique complete support and two compatible coefficient matrices. The same experiments admit a stable feedback model with the shortcut absent. The result extends to connected graphs with repeated branching and merging. Finite response-column alternatives yield exact bounded-parent compatibility decisions through Horn propagation and partial-target completion. Certified ancestry then supplies an observed-ancestor covariance restriction. Gaussian regression and Fieller inversion reduce inference to a scalar confidence set, and quadratic elimination retains the shared nuisance coefficients with finite-sample edge-error control. In 2048 Gaussian conditions, the scalar procedure reports 127/128 positive edges at gap 1/4 and 524288 observations per cell. Its decisions agree with the joint outer-polynomial reference on the same inputs in all 2022 qualified conditions, with zero reports in 512 null conditions. Exact 16-, 32- and 64-node evaluations verify support and target-coverage conclusions. A root-target characterization connects the identifying information to intervention design.
Qingyue Zhao, Qiwei Di, Jiafan He, Heyang Zhao, Kaixuan Ji, Quanquan Gu
Thu 19 Nov
cs.LGmath.OCstat.ML
pdf html
- favorites near “Double Reinforcement Learning for Efficient Off-Policy Evaluation in Markov Decision Processes” 0.89
- your work near “Timing as an Action: Learning When to Observe and Act” 0.83
- interest Reinforcement learning, policy learning and off-policy evaluation 0.83
Abstract
We study variance-dependent reinforcement learning in episodic time-inhomogeneous linear Markov decision processes (MDPs) with feature dimension $d$, horizon $H$, and $K$ episodes. While recent works have established variance-dependent regret with $\tilde{O}(1)$ deterministic-regime overhead for linear mixture MDPs (Zhao et al., 2023) and worst-case nearly minimax optimal regret $\tilde{O}(d\sqrt{H^3 K} + d^7 H^8)$ for linear MDPs (He et al., 2023),, existing variance-aware analyses for linear MDPs (Li and Sun, 2023; Huang et al., 2023) rely on a coarse concentration step whose resulting regret guarantees retain a polynomial $\tilde{O}(K^{1/4})$ additive term even in deterministic environments. We propose VA-LSVI-UCB++, a computationally efficient algorithm that achieves the variance-dependent regret bound $\tilde{O}(d\sqrt{H\min\{\mathrm{Var}_K^*,\mathrm{Var}_K^\pi\}}+d^5H^4)$, where $\mathrm{Var}_K^*$ and $\mathrm{Var}_K^\pi$ denote the cumulative conditional variances of the optimal value function and the played policies' value functions along the realized trajectory. Our result adapts to the minimum of the two realized trajectory variances, recovers the worst-case minimax optimal leading term $\tilde{O}(d\sqrt{H^3 K})$, and collapses to $\tilde{O}(d^5 H^4)$ in deterministic environments, free of polynomial-in-$K$ dependency. At the core of our analysis is a simultaneous three-way reference-advantage decomposition for the optimistic, pessimistic, and squared value function estimators coupled with variance self-bounding and Freedman-type value-difference arguments, which tightens the pessimistic and second-moment confidence radii by a factor of $\tilde{O}(d H)$ and removes the inverse-variance-floor blowup.
Terje Mildner
Wed 18 Nov
math.STstat.TH
pdf html
- reading theory, contraction rates, rates, infinite
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.82
- interest Bayesian computation and structural estimation 0.84
Abstract
We prove almost sure contraction rates and consistency for Gibbs posterior measures on arbitrary hypothesis spaces, in general infinite dimensional, under dependent data, nonadditive and time-dependent losses and learning rates, misspecified models, and unattainable minimisers of the limiting objective. In particular, we develop a novel proof approach in contraction rate theory by studying the asymptotics of Gibbs measures through Lorentz spaces, real interpolation theory, and Laplace transforms dispensing with testing or variational formulations which are unavailable in the considered generality. As a consequence we obtain almost sure contraction rates for dominated Bayes and fractional Bayes posterior measures in the same generality, extending existing theory on misspecified Bayesian inference. Furthermore, we demonstrate our theoretical results on novel applications to Gibbs posteriors for Dirichlet process priors with a nonadditive Wasserstein metric loss function under ergodic, non-Markovian observables.
M. Romero-Madroñal, M.D. Jiménez-Gamero, B. Milošević
Fri 13 Nov
stat.ME
pdf html
- interest Adaptive data collection and anytime-valid inference 0.87
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.86
- reading ac, nonparametric, little, inferences
Abstract
Whether to base inferences on available-case (AC) or complete-case (CC) estimators is a long-standing practical dilemma when data are missing. This issue has been investigated for point estimation in specific problems, but hypothesis testing has received comparatively little attention. The tools needed to address this problem are developed in a fully nonparametric framework for hypotheses concerning parameters that can be estimated through functions of vectors of $U$-statistics, under a MCAR missingness mechanism. The joint asymptotic normality of the AC estimators is established, permitting availability probabilities to vary with the sample size. The limiting covariance matrix factorizes into the complete-data covariance matrix and a matrix determined solely by the availability pattern, with both components admitting consistent estimators. This fact enables the construction of a Wald-type test statistic. The CC approach arises as a special case. These results allow AC and CC based tests to be compared, although the comparison is problem specific, which limits its practical value. Rather than selecting one approach, a test based on the maximum of the two statistics is also considered, whose critical value can consistently approximated by two bootstrap procedures. In simulations, its power lies between those of the AC and CC approaches.
Rikuki Okamoto
Tue 17 Nov
math.STmath.PRstat.TH
pdf html
- favorites near “Priors for the Long Run” 0.85
- interest Macroeconometrics and time series 0.84
- reading moment, diffusion coefficients, bandwidths, estimator allows
Abstract
We propose a sine-based modification of the separating information maximum likelihood (SIML) estimator for integrated volatility and covariance under equidistant sampling with microstructure noise. The estimator allows for nonzero initial observation noise. We establish consistency under moment assumptions and asymptotic normality under first-order Malliavin regularity of the diffusion coefficients. We also construct a feasible noise-bias correction using an estimator of the unknown noise covariance. This correction permits wider spectral bandwidths for consistency and stable asymptotic normality.
Aaditya Ramdas
Tue 17 Nov
math.STcs.ITmath.ITstat.MEstat.MLstat.TH
pdf html
- interest Online learning and statistical learning theory 0.88
- favorites near “Orthogonal Statistical Learning” 0.89
- reading anytime-valid, kernels, main, implementations
Abstract
For each standard Borel observation space, we construct a single anytime-valid two-sample procedure that is asymptotically log-optimal for every pair of unknown probability laws and every limiting ratio of diverging sample sizes, including zero and infinity. For independent balanced pairs, the optimal log-wealth rate is $2\JS(P,Q)$, where $\JS$ is the Jensen--Shannon divergence. Unequal sizes yield weighted Jensen--Shannon information and the one-sample limit $\log W(n_1,n_2)/n_2\to\KL(Q\|P)$ when $n_1/n_2\to\infty$. Refining finite partitions and logarithmic-loss learning with measurable characteristic kernels provide universal constructions. Our main contribution is the generality of our construction, and its connections to classical batch efficiency notions. Practical implementations allow continuous classifier fine-tuning, as demonstrated on finite alphabets, Gaussian data, and handwritten digits.
Howard Kim
Fri 13 Nov
stat.APcs.CLcs.CY
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.84
- interest Causal inference and identification in econometrics 0.82
- reading literature, nominal, did, price
Abstract
Survey estimates for population segments, such as media use by sex and age, rest on few respondents per cell, and a growing literature offers synthetic respondents generated by large language models to fill the gap. That literature has not tested them against the small-area estimators a methodologist would otherwise use. We ran ten estimators, preregistered, on identical household-clustered splits of a Korean media panel across five calibration budgets and three holdout regimes. A multilevel partial-pooling model of the MRP family led every calibration-only estimator at every budget, matching at one hundred respondents what direct estimation reaches with roughly four times as many. On unobserved segments, information fixed as a per-segment prior was inert for synthetic panel and previous wave alike; attached as a bias correction it transferred, at the price of estimation noise. No registered synthetic advantage over MRP survived the paired decision rule. Entered instead as a covariate inside the multilevel model, an exploratory hybrid, the same panel improved on MRP on unobserved segments: the information survives where the registered constructions did not. Interval coverage stayed below nominal throughout and worsened with budget for partial pooling, contrary to prediction.
Sie Hendrata Dharmawan, Peter Chin
Wed 18 Nov
math.STstat.TH
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.83
- favorites near “Optimal Decision Rules for Weak GMM” 0.89
- interest Semiparametric inference and debiased machine learning 0.84
Abstract
Mixed-membership models describe each unit as a graded blend of latent types. Before the blend is interpreted, the data should be able to rule out the simpler story in which every unit belongs to a single type. We ask when they can, in a scalar graded Gaussian mixture model, in which each observation is a Beta-weighted convex combination of independent Gaussian draws from two types, the mechanism behind functional mixed-membership models and the normal compositional model. The model is identifiable, unequal variances included, but the identifying information sits in the far tails of the marginal. For the symmetric model we give a closed-form hard-membership mixture that matches the graded marginal in five cumulants, and a Le Cam bound turns it into a certified sample size below which no test separates graded from hard membership with summed error under one half. The threshold is governed by the concentration $c$ relative to $c_0\approx d^2$, where $d$ is the type half-separation in within-type standard deviations. At $d=3$ it is about 2,600 observations at $c=3$ and 77,000 at $c=6$. Most of the difficulty comes from learning $c$. On 100 simulated datasets with graded membership, a correctly specified posterior under the default priors of a published Bayesian sampler for this model class reports hard membership every time, a weak prior gives intervals for the concentration spanning orders of magnitude, and the sampler's own answers depend on where it starts.
Boyuan Ning
Tue 17 Nov
math.STstat.TH
pdf html
- reading local asymptotic, initial value, argument, scale
- favorites near “Priors for the Long Run” 0.83
- interest Adaptive data collection and anytime-valid inference 0.84
Abstract
For the ergodic Cox--Ingersoll--Ross (CIR) diffusion with known diffusion scale and a fixed initial value, we prove local asymptotic normality (LAN) of the exactly sampled experiment for the drift parameters whenever the Feller index exceeds $1$, assuming only a vanishing mesh and a diverging observation horizon. This improves the previously known condition that the index exceed $(11+\sqrt{89})/2$. The proof cuts off the continuous score near $0$, recovers the remainder by exact conditional centering on each sampling interval, and transfers LAN by a conditional-likelihood argument.
Amir Ali Farzin, Philipp Braun, Iman Shames
Wed 18 Nov
stat.MLcs.LG
pdf html
- favorites near “Hierarchies of Relaxations for Online Prediction Problems with Evolving Constraints” 0.86
- interest Reinforcement learning, policy learning and off-policy evaluation 0.82
- your work near “Efficient Online Estimation of Causal Effects by Deciding What to Observe” 0.82
Abstract
We consider online submodular minimisation under bandit feedback, where at each round a decision maker plays a subset and observes only its cost, while the cost functions, and hence their minimiser, change over time. Existing bandit methods only guarantee static regret, which compares against a single fixed subset and is a weak benchmark when the minimiser moves. This paper provides, to the best of our knowledge, the first method with guarantees for tracking a moving minimiser in this setting. We propose a zeroth-order method built on a two-point oracle for the Lov\'asz extension whose two evaluations are both subsets that can be played. The perturbed and base points are rounded by a common random threshold, and drawing the smoothing direction from the unit sphere makes the base evaluation a zero-mean control variate, so the oracle is an unbiased estimator of the smoothed gradient. We establish dynamic regret bounds of order $O\big((1+\sqrt{P^\ast_N})N^{3/4}\big)$ in general and $O\big((1+\sqrt{P^\ast_N})N^{2/3}\big)$ under a sufficiently small drift, where $P^\ast_N$ is the path length of the minimiser sequence, together with the corresponding $O(N^{3/4})$ and $O(N^{2/3})$ static regret bounds. Numerically, the method attains lower regret than the algorithm of (Hazan and Kale (2012)), with the advantage growing with the path length.
Liping Zhu, Zhongming Xie
Fri 13 Nov
math.PRmath.STstat.MEstat.TH
pdf html
- interest Causal inference and identification in econometrics 0.85
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.79
- favorites near “Advances in Nowcasting Economic Activity: Secular Trends, Large Shocks and New Data” 0.80
Abstract
Sample distance correlation can remain large between independent random walks. For centred walks with independent, identically distributed increments of finite, positive variance, we prove convergence in distribution to the distance correlation of the joint occupation measure of two independent Brownian paths. This universal limit lies strictly between zero and one almost surely and has full support on the unit interval. We calculate the expected squared distance covariance of the limiting measure and show that ordinary permutation calibration rejects independence with probability tending to one at every fixed nominal level. Gaussian simulations illustrate the persistent dispersion of random-walk correlations and the failure of permutation calibration.
María Durbán, Vanesa Guerrero, Cristina Molero-Río
Tue 17 Nov
stat.MEmath.OCmath.STstat.TH
pdf html
- interest Semiparametric inference and debiased machine learning 0.88
- reading functional, functional regression, functions, function-on-function
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.82
Abstract
Standard regression methods do not always guarantee that the estimated model reflects prior knowledge about how predictors relate to the response, which motivates constrained regression. When the effects to be estimated are smooth nonlinear functions, a natural family of constraints is that of shape constraints, such as sign, monotonicity, or curvature, which encode qualitative properties that these functions are often expected to satisfy. While shape-constrained estimation of smooth functions is well established in classical regression with scalar covariates and response, its extension to functional regression, where the coefficients are themselves functions, remains largely unexplored. In this paper, we propose the functional shape-constrained conic P-spline estimator (FunSCOPE), a unified framework for shape-constrained functional linear regression that covers scalar-on-function, function-on-scalar, and function-on-function models, and in which any combination of the shapes above can be imposed on the functional regression coefficient. Building on a P-spline approach, the desired shapes are expressed as conditions on the scalar regression coefficients, which leads to an estimation problem formulated as a conic optimization problem. For univariate coefficients, as in scalar-on-function, function-on-scalar, and a particular case of the function-on-function regression, the concurrent case, these conditions are both necessary and sufficient, in contrast to existing approaches, which provide only sufficient conditions; for bivariate coefficients, as in general function-on-function models, they are neither necessary nor sufficient. Numerical experiments show that FunSCOPE outperforms competing methods in terms of estimation error and computational time.
Catarina P. Loureiro, M. Rosário Oliveira, Paula Brito
Fri 13 Nov
stat.COstat.ME
pdf html
- interest Semiparametric inference and debiased machine learning 0.88
- favorites near “Distributional Random Forests: Heterogeneity Adjustment and Multivariate Distributional Regression” 0.87
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.77
Abstract
AIDA is an R package that provides tools for the analysis of interval-valued data, including construction, visualization, robust estimation, and explainable outlier detection. The package is built around the intData class and is designed to support methodological research and applied workflows involving interval-valued data. AIDA implements methods for symbolic covariance estimation based on the Mallows distance, robust estimation via the Interval Minimum Covariance Determinant estimator, and explainable outlier detection using Shapley values. The package also includes visualization tools to support interpretation of the results and several interval-valued datasets.
Gianluigi Pillonetto, Mauro Bisiacco
Thu 19 Nov
stat.MLcs.LGcs.SYeess.SY
pdf html
- reading kernels, hilbert, gains, kernel-based
- interest Kernels, operators and functional data 0.88
- favorites near “Operator-valued Kernels for Learning from Functional Response Data” 0.86
Abstract
Kernel-based regularization for linear system identification uses positive semidefinite kernels both to define reproducing kernel Hilbert spaces (RKHSs) and as covariance models for stochastic impulse responses. Sharp kernel conditions are known for RKHS stability, while almost-sure BIBO stability under Gaussian priors requires an even more restrictive covariance condition. These restrictions, however, need not persist beyond Gaussianity, where the covariance no longer determines the full probability law. We introduce the notion of a stability-compatible kernel, namely a covariance kernel admitting at least one stochastic realization with almost-surely absolutely summable sample paths. We show that stability-compatibility holds if and only if the stable elements are dense in the associated RKHS, a condition strictly weaker than RKHS stability, and derive practical sufficient conditions, including a spectral criterion for stationary kernels. We then construct non-Gaussian priors that preserve a prescribed covariance while enforcing almost-sure stability. In particular, this yields a stable prior based on the stationary radial-basis-function (RBF) kernel, separating covariance structure from the mechanism enforcing stability. Numerical experiments show substantial estimation gains for several important classes of impulse responses, with particularly pronounced advantages when relevant dynamics occur at late lags or in separated temporal regions.
Thiyanga S. Talagala
Tue 17 Nov
stat.CO
pdf html
- reading series, appear, year, time series
- favorites near “Advances in Nowcasting Economic Activity: Secular Trends, Large Shocks and New Data” 0.82
- interest Macroeconometrics and time series 0.85
Abstract
`tsspiral` is an open-source R package that introduces a circular approach to time series visualisation using polar coordinates. We refer to this circular representation of time series data as a spiral. `tsspiral` provides a collection of `ggplot2` geometries for creating a spiral representation of time series data. In our approach, we arrange observations around a circle. Each year is represented by a circle, with the angular position corresponding to the day of the year, and successive years appear as concentric rings. This arrangement facilitates the exploration of periodicity, seasonality, and recurring temporal patterns, particularly in long time series. The package also supports temporal hierarchical visualisations to compare patterns across time scales. By bringing polar-coordinate-based visualisation into routine time series analysis, `tsspiral` offers a practical tool for exploring temporal patterns and communicating complex time-dependent structures. Rather than treating spiral plots as standalone graphics, our approach brings them into the grammar of graphics framework, enabling users to explore temporal patterns through a broader range of visual encodings.
Mansi Sood, Devavrat Shah
Fri 13 Nov
stat.MLcs.LGeess.SP
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.83
- interest Semiparametric inference and debiased machine learning 0.86
- favorites near “The Minimax Regret of Sequential Probability Assignment, Contextual Shtarkov Sums, and Contextual Normalized Maximum Likelihood” 0.86
Abstract
Motivated by causal inference and physical computing systems, we consider the problem of learning mixture of two bounded-support exponential-family distributions with a shared interaction parameter matrix and mixture component-specific linear parameter vectors. The unknown sequence of component labels belongs to a known structured class. We propose an estimator based on a convex loss that avoids the log-partition function; for each fixed label assignment, estimation is a convex optimization problem. We show that the conditional population loss uniquely identifies the interaction matrix and the linear parameter attached to each sample. For a global empirical minimizer, we establish finite-sample bounds on the interaction error in Frobenius norm and the average squared linear-parameter error. The bounds depend on the logarithm of the number of admissible label assignments, and, for fixed dimension and model bounds, imply consistency when this complexity grows sublinearly with sample size. Additional occupancy and separation conditions yield component and label recovery.
Winston Chou, Mårten Schultzberg, Lo-Hua Yuan, Wenjing Zheng, Avi Feller, David Holtz, et al.
Fri 13 Nov
cs.AIstat.AP
pdf html
- reading experimentation, science, practice, causal
- favorites near “Data Science at the Singularity” 0.88
- interest Language models as statistical objects 0.80
Abstract
Artificial intelligence (AI) is quickly expanding the capacity of organizations to produce and analyze data. We argue that this expansion shifts the leverage of data scientists from performing individual analyses to assessing the credibility and importance of findings and determining when evidence generalizes across contexts. Drawing on a roundtable of academic researchers and industry practitioners, we examine three ways in which data scientists who work on causal inference and experimentation can help organizations turn more analytical capacity into better decisions: preserving and accumulating the learning value of experimentation beyond individual tests; choosing and combining experimental, observational, and synthetic evidence; and ensuring evaluation keeps pace with AI as its capabilities and uses evolve. In each area, we distinguish established practice from open questions. We conclude with implications for the training and practice of data science and for organizations.
Anjana Grandhi, Wenge Guo, Ji Meng Loh, Sanat Sarkar
Tue 17 Nov
stat.ME
pdf html
- interest Adaptive data collection and anytime-valid inference 0.88
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.85
- favorites near “The Minimax Regret of Sequential Probability Assignment, Contextual Shtarkov Sums, and Contextual Normalized Maximum Likelihood” 0.84
Abstract
We study false discovery rate (FDR) control for a fixed family of hypotheses tested sequentially in a prespecified order. Each decision uses only the current and preceding $p$-values, is not subsequently revised, and permits continued testing after non-rejections. We distinguish generalized fixed-sequence procedures based on preceding rejection counts from fallback-type procedures that additionally use recursively transferred testing levels. For rejection-count procedures, we study a position-dependent calibration under independence and a linear rule calibrated through null marginal distribution functions under positive left-tail dependence under the null. We clarify their connections to existing fixed-sequence and online FDR methods. The fallback construction combines the total number of preceding discoveries with level transfer along consecutive rejections, allowing thresholds to reflect the arrangement of earlier outcomes. We establish its FDR control under history-based positive regression dependence and, with a position-specific correction, under arbitrary dependence. For all four constructions, the guarantees extend to every deterministic testing time within the planned horizon. Gaussian simulations assess the effects of signal placement, dependence, and allocation weights. They demonstrate power gains over selected benchmarks while identifying the conservativeness of the arbitrary-dependence correction.
Trisha Lawrence
Tue 17 Nov
stat.AP
pdf html
- interest Computational methods for heterogeneous agent and dynamic economic models 0.80
- reading market, markets, programming, success
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.81
Abstract
Electricity generation has become a remunerative avenue for new investments, particularly in the area of renewable energy resources. This paper introduces a framework for integrating statistical methods with optimization modeling to improve decision support in electricity generation. The analysis begins by evaluating the success of mini-grid projects installed across the globe using both a probit and a Bayesian hierarchical model to estimate the probability of project success. Furthermore, we provide a multi-resource optimization framework for determining the optimal profit in electricity generation while reducing carbon emissions in production facilities. We also analyze the impact of a battery energy storage system and bidding behavior in the Alberta electricity market by constructing a mixed-integer self-scheduling model that determines optimal charging and discharging strategies. Numerical experiments show how storage systems can enhance energy arbitrage profitability and influence market clearing prices. Finally, we propose a Stochastic Dynamic Programming (SDP) model to maximize profit in electricity generation while accounting for price volatility. Integrating statistical methods with optimization-based modeling reveals great value in achieving more effective, profitable, and clean energy systems.
Even He, Zaizai Yan
Fri 13 Nov
stat.MEstat.ML
pdf html
- interest Semiparametric inference and debiased machine learning 0.89
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.86
- favorites near “Deep Learning for Individual Heterogeneity” 0.81
Abstract
Cross-fitting is used to test whether a ranking learned in a randomized experiment separates units that benefit from treatment from those that do not: each fold is ranked by a model fitted to the other folds, and the standardized difference in estimated effects between units ranked high and low is compared with a normal distribution. Under the null hypothesis of no heterogeneity the ranking is determined by noise, and because each fold helps to rank the others the fold estimates are dependent. We show that this dependence changes the shape of the null distribution, not just its variance. For least squares on one covariate the limit is bimodal for two folds and skewed for many, no variance correction restores the level, and the model is a counterexample to a published central limit theorem. With many covariates the limit approaches normality, but its variance stays inflated. Removing the non-normal component of the statistic removes its power, so we keep the statistic and change its reference distribution: joint calibration simulates together the moments that choose the rankings and those that evaluate them and applies the same selection to the draws; it has asymptotically exact level and requires no refitting. Applications to clinical and marketing trials, together with random treatment reassignments on their covariates, demonstrate that the standard test substantially inflates Type I error and can produce spurious rejections that joint calibration avoids.
Kunqi Huang, Lican Kang, Hongrui Shu, Jincheng Ying
Wed 18 Nov
stat.MLcs.LG
pdf html
- favorites near “Double Reinforcement Learning for Efficient Off-Policy Evaluation in Markov Decision Processes” 0.86
- interest Reinforcement learning, policy learning and off-policy evaluation 0.85
- your work near “Timing as an Action: Learning When to Observe and Act” 0.80
Abstract
Learning reliable control policies requires accounting for uncertainty in transition dynamics and errors introduced by function estimation and action generation. We develop an actor--critic method for discounted robust Markov decision processes with continuous states and actions and an unknown nominal transition law. We consider transition uncertainty sets defined by Wasserstein distance, total variation, Kullback--Leibler divergence, or Pearson chi-square divergence. A conditional diffusion model is fitted once to offline transitions. At each subsequent iteration, we estimate the robust action-value function from current-policy observations using a shared-target minimax criterion, then update a generative actor through exponentially weighted score matching. Regularization preserves exploration and provides the density regularity needed for conditional distribution estimation. Under coverage and regularity conditions, we establish finite-sample bounds for critic estimation, the learned conditional action law, and the robust return of the implemented policy. The bounds distinguish critic estimation error, uncertainty in the nominal transition model, and policy regularization, and characterize how the four robustness radii affect the resulting guarantees.
Haoge Chang, Kota Saito, Mingxi Zhu
Mon 16 Nov
econ.EMecon.TH
pdf html
- reading rational, choice, conceptual, generalize
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.80
- interest Reinforcement learning, policy learning and off-policy evaluation 0.82
Abstract
This note supplements our paper \textit{Alternative Computational Formulations for Testing Discrete-Choice Random Utility Models: Half-Space and Network-Flow Formulations} and serves two purposes: 1. We clarify the conceptual distinctions between two classes of random utility models (RUM): those in the consumption-budget setting \citep{kitamura2019nonparametric} (hereafter KS) and those in the finite-alternative setting (e.g., Chapter~7 of \cite{chambers2016revealed}). In particular, we show (Theorem 1) that each rational type, as defined in KS, can be obtained as a coordinate projection of the deterministic choice vector induced by a strict preference ranking that satisfies what we call a patch-monotonicity condition. 2. We generalize the network-flow characterization for finite-alternative RUM in \cite{fiorini2004short} to the consumption-budget setting (Theorem 2). Specifically, we show that a vector of choice probabilities in the consumption-budget setting is rationalizable in the KS sense if and only if it is the image of a unit flow on a multi-arc powerset digraph under a fixed linear map. The key distinction is that our digraph allows multiple arcs between the same pair of nodes, whereas the digraph in Fiorini has at most one arc between any pair of nodes.
Omar Kassi, Bernhard Stankewitz, Botond Szabo
Fri 13 Nov
math.STstat.MLstat.TH
pdf html
- reading processes, kernels, results standard, commonly encountered
- interest Kernels, operators and functional data 0.84
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.79
Abstract
Gaussian process regression models are widely used in modern statistics and machine learning due to their flexibility, interpretability, built-in uncertainty quantification, and strong theoretical foundations. However, training Gaussian processes (GPs) has a computational cost of $O(n^3)$, while the memory and prediction costs are also $O(n^2)$. For large datasets commonly encountered in practice, this becomes infeasible and necessitates the use of approximate posterior methods. In this work, we propose a class of approximations based on sparse covariance matrices, where the kernel matrix is sparsified according to the spatial proximity of the design points. We analyze the accuracy of the resulting approximate posterior distribution and show that the proposed approach preserves optimal statistical inference properties, provided that the dependencies among a sufficient number of neighboring points for each feature are retained. We derive explicit sufficient and necessary conditions on the number of neighbors required in a general setting and apply these results to standard kernels, including the squared exponential and Mat\'ern kernels. Our theoretical findings are supported by numerical experiments.
Benjamin D. Youngman, Ayu Shabrina
Mon 16 Nov
stat.MEstat.CO
pdf html
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.78
- interest Macroeconometrics and time series 0.86
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.81
Abstract
We introduce \textsf{evgmrf}, an \textsf{R} package for fitting spatially varying extreme value models using Gaussian Markov random fields (GMRFs). The package supports the generalized extreme value distribution for block maxima, the generalized Pareto distribution for threshold exceedances, the Poisson--GPD point-process model, and the $r$-largest order-statistics model. Distributional parameters may vary over regular spatial grids or arbitrary regions that can be represented by a valid adjacency matrix, with optional fixed effects and covariate dependence. Spatial structure is represented through intrinsic conditional autoregressive, proper conditional autoregressive, and Besag--York--Molli\'{e} 2 models, which are extended with non-Gaussian random-effect distributions. Estimation uses penalized likelihood with smoothing or dependence hyper-parameters estimated using a Laplace approximation, exploiting sparse precision matrices for computational efficiency. The package provides methods for model summaries, spatial visualization, prediction on link and response scales, estimation of extreme quantiles, and uncertainty quantification in the form of approximate standard error estimates, or via simulations from parameter sampling distributions. Examples based on gridded and county-level precipitation data illustrate the use of the package for spatial modelling of environmental extremes. \textsf{evgmrf} provides a unified, accessible framework for extreme value analyses in which distributional parameters exhibit structured spatial variation.
Deepesh Bhati, Bruno Ebner, Sakshi Khandelwal
Wed 18 Nov
math.STstat.TH
pdf html
- interest Semiparametric inference and debiased machine learning 0.88
- your work near “Perturbation Methods for Incomplete Markets Economies” 0.83
- favorites near “Revisiting the Phillips and Beveridge Curves: Insights from the 2020s Inflation Surge” 0.82
Abstract
A Stein--Laplace characterization yields two closed-form goodness-of-fit statistics for the Pareto type-I family with unknown shape and known lower endpoint. The signed integral statistic $T_n$ has a standard normal limit after studentization, but explicit non-Pareto alternatives show that it is not omnibus consistent. The integrated squared discrepancy $S_n$ yields a consistent test against every fixed non-Pareto alternative on the specified support with a finite, positive logarithmic moment. We derive the weighted chi-square null limit of $nS_n$ and give conditions for an asymptotically valid parametric-bootstrap and plug-in covariance-operator method. The latter uses Nystr\"om eigenvalue approximation to avoid repeated sample generation and parameter estimation. A large Monte Carlo comparison study shows competitive performance relative to existing procedures in the literature. In the paired study, the two calibrations have similar empirical power, with slightly higher small-sample rejection rates for the bootstrap in some settings. In the timing experiment, covariance-operator calibration is approximately $28$--$37$ times faster at $n=200$. An exploratory threshold-sensitivity analysis of Danish fire-insurance losses illustrates the dependence of Pareto goodness-of-fit conclusions on the specified tail region.
Miles E. Lopes, Ricardo Masini
Mon 16 Nov
math.STmath.PRstat.TH
pdf html
- reading interpolation, question, derivatives, special
- interest Adaptive data collection and anytime-valid inference 0.84
- favorites near “Simplifying debiased inference via automatic differentiation and probabilistic programming” 0.85
Abstract
We consider Berry-Esseen bounds over rectangles for sums of $n$ random vectors in $\mathbb{R}^p$ that are i.i.d. A longstanding open question in high-dimensional probability and statistics is whether bounds of order $1/\sqrt n$ are attainable up to factors of $\log(pn)$ when the population covariance matrix is unrestricted. We answer this question affirmatively, provided that the random vectors have sub-Gaussian entries with non-zero variance. In addition, we provide analogous results for bootstrap approximation and Gaussian comparison, which make use of a special discrepancy measure between covariance matrices that enables improvement with respect to prior work. Taking an approach based on interpolation, the novel aspects of the proofs deal with bounding the interpolation derivatives in a way that is not sensitive to the covariance structure.
Sreejeet Maity, Sukanya Modak, Leonardo F. Toso
Wed 18 Nov
cs.LGstat.ML
pdf html
- reading state, history, long-run, state space
- your work near “Timing as an Action: Learning When to Observe and Act” 0.82
- favorites near “Double Reinforcement Learning for Efficient Off-Policy Evaluation in Markov Decision Processes” 0.83
Abstract
We study the discounted reinforcement learning (RL) problem in a setting where the underlying environment changes between deployments. Each deployment provides $H$ observations from a fixed MDP. The reward function and transition kernel are fixed within each deployment. Across deployments, at most an $\epsilon >0$ fraction of the reward and transition kernel entries may change. In this setting, learning each model to deploy from scratch may discard useful observations from past deployments, whereas pooling all past observations can bias the estimates of entries that have already changed. We then propose an algorithm that for each state and action pair, selects how much history to reuse to estimate the reward and probabilities of the next state. We start with observations from the current deployment and add earlier deployment observations one at a time, the most recent first. After each addition, we check whether a single average reward agrees with the estimates from all histories considered until that point. We apply the same check to the distribution of next states. For each estimate, we leverage the last history that passes these aforementioned checks. The optimal control policy is then computed on top of the fitted MDP. We characterize the average gap between the optimal return and the return of the learner's policies over a finite number of deployments. We demonstrate that when $0< |S|\epsilon \leq 2$, with $|S|$ being the cardinality of the state space, our algorithm needs only the observations from the most recent $O(1/\epsilon)$ deployments, and achieves a long-run average gap of at most $\widetilde{O}(\sqrt{\epsilon/H})$.
Tristan Luca Saidi, Arun Kuchibhotla
Wed 18 Nov
stat.MEstat.ML
pdf html
- interest Semiparametric inference and debiased machine learning 0.89
- favorites near “Eigendecompositions of Transfer Operators in Reproducing Kernel Hilbert Spaces” 0.84
- your work near “Automated Solution of Heterogeneous Agent Models” 0.82
Abstract
In this work we propose and analyze a procedure for estimating a generalized single-index regression model. This generalized single-index regression framework models the conditional mean of the responses $Y$ as a function of the covariates $X$ where the mapping is assumed to have the form $f \circ g$, where $g$ is $\beta$-H\"older with $\beta \geq 2$ and $f$ can be a discontinuous function of bounded variation. We propose a general procedure that operates under minimal assumptions about the distribution of covariates $X$, requiring only boundedness, sufficient smoothness and absolute continuity with respect to Lebesgue. Our procedure jointly estimates the gradient vector $\hat v(x) \approx \nabla g(x)$ and the conditional mean $\mu(x) = f(g(x))$ by solving a localized weighted linear least squares problem. We derive central limit theorems for gradient vector estimates $\hat{v}_x$ and regression estimates $\hat{\mu}(x)$, and we derive rates of convergence for our approach, illustrating that it indeed achieves a stochastic convergence rate matching the minimax mean squared error rate when $\beta = 2$ and $f$ has a locally Lipschitz derivative. Unlike prior approaches our velocity estimation results accommodate link functions in the local bounded-variation class, allowing for nonsmooth and even discontinuous links while retaining consistency and asymptotic normality at almost every inference point. Finally, we instantiate this model in the situation where $g = \Pi$ is a closest point projection onto a compact and regular embedded $C^3$ curve $\gamma$, a setting recently studied in the literature. We also supplement our theoretical results with experiments on real and synthetic data to demonstrate the utility of our procedure.
Stefan Oberdörster
Tue 17 Nov
math.PRcs.LGstat.COstat.ML
pdf html
- interest Computational methods for heterogeneous agent and dynamic economic models 0.81
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.81
- reading equilibrium, hamiltonian monte, technology, hamiltonian
Abstract
Hamiltonian Monte Carlo, since its invention and subsequent rise to widespread use, has been motivated by its ability to accelerate convergence to equilibrium compared to diffusive methods, both in terms of dependence on condition number and dimension. In the present work, we establish this acceleration simultaneously in both aspects for critically damped Metropolis-adjusted kinetic Langevin with persistent uniforms. More precisely, for target measures $\mu(dx)\propto e^{-U(x)}dx$ with $m$-strongly convex, $L$-gradient Lipschitz, and $L_H$-Frobenius Hessian Lipschitz potentials, we show that the method mixes to $\varepsilon$-accuracy in total variation using $\mathcal O\bigl((\kappa^{1/2}+L_H^{1/2}m^{-3/4})\,d^{1/4}\bigr)$ gradient evaluations up to logarithmic factors, both uniformly over starting points in domains of high equilibrium probability and from warm starts. The proof incorporates ideas from previous studies by Bou-Rabee and Chen et al. that show either aspect of acceleration separately, into novel coupling technology. This new coupling technology for general Markov chains extends earlier frameworks aimed at capturing fast local equilibration in the presence of slow global mixing. Specifically, via a weak-$(1,1)$ maximal inequality that follows from Hopf's maximal ergodic theorem, we uncover a stability mechanism in high-probability domains with respect to equilibrium. This enables straightforward localization to naturally arising domains of fast local mixing, and seamlessly fits broad local frameworks via Dobrushin contraction, Wasserstein contraction and local Harris' theorems. We develop the Dobrushin formulation.
Alessia Berarducci, Melania Lembo, Veronica Vinciotti, Ernst C. Wit
Tue 17 Nov
stat.ME
pdf html
- interest Semiparametric inference and debiased machine learning 0.88
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.85
- reading causal, generalize, make, invariance
Abstract
Prediction is an important use of a statistical model. However, predictive systems typically struggle when the environment, in which they were trained, changes. Robust prediction that can deal with distributional shifts has become an active research area. Existing approaches tend to make strong assumptions in order to achieve prediction guarantees. In this paper, we only require finite second moments in the training data to obtain sharp prediction guarantees. We generalize causal regularization to a framework that uses two environments and measures the differences between them through a generalized risk difference induced by changes in second-order moments of the covariates and response. By defining a sieve of out-of-sample distributions, we derive a worst-risk decomposition over increasing classes of environments and obtain a closed-form estimator that interpolates between OLS and the regularization limit solution. Causal interpretation is recovered under the inner-product invariance condition and when the conditional mean of the target given its causal parents remains invariant across environments. Our results show that the proposed estimator can achieve prediction stability over a broad class of perturbed environments, even when trained on a limited set. The framework separates two goals: robustness as a stable prediction under weak assumptions and causality as a stronger property obtained with additional invariance structure.
Valerie R. Fu, Brayden Lu, Lu Tian
Tue 17 Nov
cs.LGstat.ML
pdf html
- interest Semiparametric inference and debiased machine learning 0.86
- reading structural, theory, preferences, consistent
- favorites near “Simplifying debiased inference via automatic differentiation and probabilistic programming” 0.85
Abstract
Stagewise algorithms and sparsity-inducing regularization are two influential approaches to high-dimensional modeling. Existing stagewise theory, however, largely centers on convex regularization, whereas stronger sparsity and other structural preferences are often pursued through fixed-level nonconvex optimization rather than generating an entire adaptive solution path. We develop stagewise adaptive boosting, a general framework that builds a monotone path by selecting the update consistent with structural preference. The framework can encourage sparsity along the path itself and also admits extensions that promote other properties of the solution, including group selection, similarity across models for multiple outcomes, and known priorities or prerequisite relations among predictors. We study its theoretical property as well as finite sample performance using simulations and real-data analyses. Across these studies, the proposed pathwise approach often achieves predictive performance comparable to commonly used baselines while selecting substantially smaller models. These results suggest that adaptive stagewise learning provides a practical route to stronger sparsity and richer structural control without repeatedly solving separate weighted or nonconvex optimization problems.
Mingrui Liu
Wed 18 Nov
cs.LGmath.STstat.MLstat.TH
pdf html
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.82
- favorites near “Hierarchies of Relaxations for Online Prediction Problems with Evolving Constraints” 0.82
- reading online, sequential, hindsight, expected cost
Abstract
Conformal prediction provides distribution-free uncertainty quantification, but in sequential deployment, standard coverage guarantees can still allow substantial fluctuations in the number of miscoverages over time. We study an online setting in which prediction sets are produced one round at a time and the cumulative number of miscoverages must stay within a prescribed budget throughout deployment.Such a hard coverage guarantee can be enforced by enlarging prediction sets, but how much additional prediction-set size is fundamentally necessary remains unclear. We characterize this unavoidable efficiency cost over $T$ prediction rounds. Remarkably, the cost persists even with i.i.d.\ data, a known score distribution, and perfect knowledge of the population-optimal prediction set. Under suitable regularity conditions relating coverage and prediction-set size, we determine the minimum cumulative excess prediction-set size required by any online method satisfying the hard coverage constraint. This minimum cost is $\Theta(\log T)$ for expected cost at a fixed time known in advance, $\Theta(T^{1/3})$ for the largest expected cost over all fixed times up to $T$, and $\Theta(\sqrt{T})$ for the expected worst realized cost when the evaluation time is chosen in hindsight. We further show that these rates can be attained using past scores, without knowing the score distribution. Thus, even perfect statistical knowledge does not make sequential reliability free.
Ghislain Geniaux
Fri 13 Nov
stat.ME
pdf html
- favorites near “Priors for the Long Run” 0.83
- interest Causal inference and identification in econometrics 0.84
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.81
Abstract
This paper develops a boosting-based framework for estimating nonlinear spatial autoregressive models (SAR and SEM). Functional gradient descent boosting with spline base learners provides automatic selection of covariates and functional forms, while the spatial parameter is estimated by an extension of the closed-form, determinant-free estimator (CFE) of Smirnov (2020): we show that it remains consistent, with the same asymptotic variance formula, for any additive estimator admitting a smoothing-matrix representation, and we prove that the underlying quasi-score is Neyman-orthogonal to the nonlinear nuisance, which makes the flexible, regularized first stage safe. This flexibility, however, exposes the estimator to a spatial confounding problem that must be solved for the framework to deliver structural estimates. The geoadditive smooth of the coordinates, s(u, v), a default ingredient of flexible spatial specifications, competes with the autoregressive term for the same low-frequency spatial variation, and purely predictive tuning resolves the competition in favour of the surface, biasing the spatial parameter -the geoadditive trap. We therefore equip the estimator with targeted safeguards: a spectral identification diagnostic, a spectral penalty on the surface, spatial cross-fitting, stopping iterations selected by spatial cross-validation, and heteroskedasticity-and spatial-correlation-robust sandwich variances. Monte Carlo experiments, including a factorial design that generates the trap, show that the safeguarded estimators restore test size and confidence-interval coverage while matching maximum likelihood accuracy at a fraction of its computational cost. An empirical application to the Lucas County, Ohio, housing market illustrates the workflow.
Maya B. Mathur, Willi Zhang, Shaun Seaman, Rohit Bhattacharya, Ilya Shpitser
Wed 18 Nov
stat.MEmath.STstat.TH
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.88
- interest Causal inference and identification in econometrics 0.82
- reading income, causal, burden, auxiliary
Abstract
The missingness-at-random (MAR) assumption underlying many missing-data methods describes abstract statistical independences, whereas practitioners reason more naturally about the causal mechanisms that generate missingness. To bridge this gap, we provide a mechanistic characterization of MAR in terms of paths in a directed acyclic graph (DAG) modified to allow certain edges to cease to operate when an incomplete variable is missing, a property termed "context-specific causal irrelevance (CSCI)". Under a faithfulness condition, MAR holds if and only if the modified DAG contains no MAR-violating paths. Without CSCI, this condition reduces to a previously established d-separation condition, VMAR; thus, MAR without VMAR requires specific kinds of CSCI, as can occur with monotone dropout. These results shift the burden on investigators from passively invoking MAR to justifying, on substantive grounds, the specific mechanisms it requires. We then consider identification of a target law p(v) of variables V by a MAR-assuming analysis, e.g., multiple imputation, that includes complete auxiliary variables Z (variables not in V). Barring a perfect cancellation, such an analysis recovers p(v) if and only if MAR holds with respect to V and Z jointly. We show that if there exists any set of auxiliaries such that (V, Z) obeys MAR, then the complete candidate auxiliaries that are ancestors of the target variables or their missingness indicators form a sufficient set. Contrary to standard recommendations, not all candidate auxiliaries should be used: even complete auxiliaries can introduce MAR violations that would not otherwise occur. We illustrate these results with simulations and an analysis of sexual identity change in the Stockholm Public Health Cohort, where our results indicate that a candidate auxiliary, income, should be excluded; including it attenuated estimates toward the null.
Sihan Wu, Yifan Cui
Tue 17 Nov
math.STstat.MEstat.MLstat.TH
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.81
- reading nuisance, equations, treatment, units
- favorites near “Bayesian Semiparametric Model for Sequential Treatment Decisions with Informative Timing” 0.86
Abstract
This paper considers estimation and inference for the average dose-response function under right-censored treatment duration and missing outcomes for censored units. Using the duration lower bounds provided by censored records, we construct a novel multiply robust estimator through augmented local estimating equations. Under suitable regularity conditions, the estimator remains consistent when any two of the outcome regression, treatment density, and censoring survival function are consistently estimated. We establish its convergence rate and asymptotic distribution within a cross-fitting framework under appropriate conditions on nuisance estimation. To address local smoothing bias that can distort confidence interval coverage, we further develop a debiased estimator that supports asymptotically valid pointwise inference without undersmoothing. Under additional uniform conditions, we construct simultaneous confidence bands using a Gaussian multiplier bootstrap. Simulation studies evaluate estimation accuracy, the validity of pointwise and uniform inference, and robustness to nuisance model misspecification. A semi-synthetic study calibrated to NHANES illustrates the proposed methods in a setting with a known dose-response function.
David Köhler, Matthias Schmid
Tue 17 Nov
stat.MLcs.LGstat.ME
pdf html
- interest Language models as statistical objects 0.87
- favorites near “Fit without fear: remarkable mathematical phenomena of deep learning through the prism of interpolation” 0.83
- your work near “Automated Solution of Heterogeneous Agent Models” 0.82
Abstract
Modern machine learning models, such as artificial neural networks, achieve high predictive performance through increasingly complex black-box architectures, limiting their use in settings where transparency, interpretability, and fairness are required. Here we present the R package ONAM, which provides a novel and practical implementation of functional decomposition, a technique to facilitate the interpretability of black-box models through structured effect representations. ONAM implements an algorithmic framework for the post-hoc decomposition of predictions into components associated with different input features. The package is designed for usability and efficiency, including functionalities for visualization as well as quantitative measures to assess interpretability. As a model-agnostic method, it can be applied to a wide range of prediction models. We describe the mathematical foundations, algorithmic design, and software architecture of ONAM. A central result of the proposed methodology is that the resulting decomposition satisfies the condition of hierarchical orthogonality, thereby guaranteeing that all components are identifiable and uniquely defined.
Xifeng Li, Shuzhen Yang
Fri 13 Nov
stat.ME
pdf html
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.82
- interest Adaptive data collection and anytime-valid inference 0.86
- favorites near “Priors for the Long Run” 0.82
Abstract
In this paper, we study generalized linear models under distribution uncertainty, where the random intercept has mean uncertainty and takes values in an unknown bounded interval. We develop a novel two-step method, named Firth-corrected Expectation-Maximization with Moving Block (FEMMB), to estimate the regression coefficients as well as the lower and upper bounds of the intercept. A Firth bias correction is incorporated into the EM algorithm to mitigate the incidental parameters problem, which is prominent in generalized linear models with many nuisance intercepts. Under mild regularity conditions, we establish the consistency and asymptotic normality of the estimated regression coefficients, and the consistency of the estimated intercept bounds. Simulation studies and a real-data application to an imbalanced credit card dataset demonstrate the effectiveness of the proposed method.
Shuchismita Sarkar, Jing Li, Michael P.B. Gallaugher, Xuwen Zhu
Mon 16 Nov
stat.ME
pdf html
- favorites near “Advances in Nowcasting Economic Activity: Secular Trends, Large Shocks and New Data” 0.80
- interest Macroeconometrics and time series 0.85
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.80
Abstract
To jointly identify temporal regimes and row-column structure in matrix-valued time series, we propose a hidden Markov model (HMM) with Gaussian co-clustering distributions. A single latent regime governs each matrix. Conditional on the regime, row and column memberships define Gaussian blocks, each with its own mean and variance. Membership probabilities and block parameters are shared across occurrences of a regime, while individual memberships may vary between time points.
Yuliya Mishura, Kostiantyn Ralchenko, Mykyta Yakovliev
Fri 13 Nov
math.STstat.TH
pdf html
- favorites near “Priors for the Long Run” 0.85
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.82
- interest Macroeconometrics and time series 0.84
Abstract
We study drift parameter estimation for a positive stochastic process with singular mean-reverting drift driven by Gaussian noise. For Riemann-Liouville fractional Brownian noise with Hurst parameter $H>1/2$, an observable fractional transformation reduces the model to a semimartingale with drift linear in the unknown parameters. Using the quasi-likelihood approach, we derive explicit estimators of both drift coefficients. For a broader class of Gaussian driving processes, we construct a strongly consistent estimator of the ratio $a/b$, which has a natural mean-reversion interpretation, and verify the required assumptions for several Gaussian models. The theoretical results are complemented by a simulation study investigating the effects of the observation horizon and the discretization step on the finite-sample accuracy of the proposed estimators.
Davide Agnoletto, Tommaso Rigon, David B. Dunson
Fri 13 Nov
stat.ME
pdf html
- interest Adaptive data collection and anytime-valid inference 0.86
- favorites near “Simplifying debiased inference via automatic differentiation and probabilistic programming” 0.87
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.81
Abstract
Predictive Bayes quantifies uncertainty by specifying a sequence of one-step-ahead predictive distributions, inducing a martingale posterior without an explicit prior. Existing parametric constructions rely on stochastic-gradient updates and have seen limited development for flexible regression. We propose a predictive Bayes framework based on recursive least squares, whose martingale posterior has an explicit representation: given future covariates, it is the law of the posterior mean plus a weighted series of independent draws from the standardized predictive distribution. The posterior mean is a penalized least squares estimator, available in closed form and invariant to data ordering, whereas the predictive distribution, for example heavy tailed, is chosen separately without changing the estimator or algorithm. We derive the exact posterior variance under a known design and an asymptotically Gaussian approximation, and obtain the martingale posterior in closed form for Gaussian predictive distributions. Posterior draws are independent by construction, with a known truncation error, and require no Markov chain Monte Carlo. The construction extends to additive, mixed and Gaussian process models. In simulations and an analysis of NHANES triglyceride data, the method attains accuracy comparable to that of a fully Bayesian model with the same response distribution at a fraction of its computational cost.
Xiao Li, Fumiyasu Komaki
Tue 17 Nov
math.STstat.TH
pdf html
- interest Semiparametric inference and debiased machine learning 0.86
- favorites near “Variational Bayes under Model Misspecification” 0.88
- reading bayesian prediction, markov, scale, density
Abstract
We consider Bayesian prediction in independent Gamma scale models with arbitrary known positive shape parameters. When the dimension is at least three, we construct a shrinkage prior toward any prespecified positive scale vector. The resulting predictive density strictly dominates the Jeffreys predictive density under Kullback--Leibler loss and is minimax. The prior is obtained by weighting the Jeffreys prior with a Markov potential. The transition is determined by the beta--gamma decomposition. The domination result is nonasymptotic.
Bilol Banerjee, Soham Sarkar
Wed 18 Nov
math.STstat.MLstat.TH
pdf html
- favorites near “Fit without fear: remarkable mathematical phenomena of deep learning through the prism of interpolation” 0.83
- reading choice, fit, locally, empirical
- interest Kernels, operators and functional data 0.84
Abstract
Deep neural networks fitted by global empirical risk minimization commit to a single model for the entire covariate space. In a regression problem, if the regression function $f_0$ is inhomogeneous --- varying rapidly in one region, while being nearly flat in another --- no single model can be appropriate everywhere, and the fit degrades where the function varies most rapidly. Here we study an alternative in which, instead of fitting the network globally, we fit it locally. For each query point $x$, a \emph{local neural network} is trained on the $k$ nearest-neighbours of $x$ alone, making the training sample and the fitted function query-dependent. We analyze the performance of the resulting estimator through a localization approach, establishing that the pointwise risk of the estimator with an appropriate choice of $k$ attains the rate $n^{-2\alpha/(2\alpha+d^{*})}$, up to logarithmic factors, when $f_0$ is locally $\alpha$-H\"older with $\alpha\in(0,1]$ and the covariates are supported on a $d^{*}$-dimensional manifold embedded in $\mathbb{R}^{d}$. A matching Le Cam lower bound shows that this rate is pointwise minimax optimal. However, the local rate-optimality requires $k$ to be calibrated to the smoothness of $f_0$, which is unknown in practice. We therefore provide a locally adaptive choice of $k$ that makes the neural network adapt to the local H\"older exponent of $f_0$ and to the intrinsic dimension simultaneously, both of which are potentially varying over the covariate domain. An extensive simulation study demonstrates the dominance of the locally adaptive neural network over a globally trained network and several state-of-the-art methods.
Eray Gençay
Fri 13 Nov
stat.APcs.LGq-fin.ST
pdf html
- favorites near “Sequential Monte Carlo Steering of Large Language Models using Probabilistic Programs” 0.83
- reading nominal, feature spaces, search, large
- interest Language models as statistical objects 0.81
Abstract
Corrections for backtest overfitting such as the deflated Sharpe ratio are applied after a search ends: the searcher maximises raw performance and a separate procedure deflates the reported winner. Autonomous searches driven by large language models perform many implicit backtests before reporting, so this paper asks what changes when the correction moves inside the loop and the acceptance threshold is indexed to the running count and dispersion of the search's own trials. The threshold then becomes an object the search can optimise against: we show analytically that a naive in-loop bar admits a variance-dilution effect, in which mediocre near-duplicate candidates lower the bar because the ledger's dispersion falls faster than its trial-count factor grows. We propose two defences, a pre-registered variance floor and a monotone high-water-mark bar, and evaluate the construction under a protocol frozen before any run, with every in-loop decision replaying exactly from its recorded trace. Across 53 audited hundred-candidate searches over two frontier models, two universes, and two feature spaces, the gated search reports "no certifiable discovery" on 36 of 37 runs, un-gated controls report a strategy on all 16, and every arm's out-of-sample confidence interval includes zero. Supporting studies find that state-based stopping reaches the same outcome after 8 to 9 trials at about a twelfth of the trial budget while an adaptive budget adds nothing, that a controlled stress test reproduces the dilution effect and both defences, and that a hundred nominal trials collapse to 6 to 8 effectively independent return streams. We scope these results to the analysed system.
Jason Beh, Pierre Del Moral, Ajay Jasra
Wed 18 Nov
stat.COcs.NAmath.NA
pdf html
- reading manifolds, state, manifold, discretization
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.81
- interest Kernels, operators and functional data 0.81
Abstract
In this article we consider the filtering of partially observed diffusions when the observations, made at discrete times, are noiseless. Each observation then confines the hidden state to a manifold defined by the data, whose type, for instance a sphere or an affine subspace, may change from one observation time to the next. In practice, the filter is approximated by particle filters, which require the diffusion to be discretized in time. Multilevel Monte Carlo reduces the cost of this discretization only if the particle approximations at consecutive discretization levels stay close, which standard particle filters do not ensure. We develop a new particle filter based on light Schr\"odinger bridges, which works between changing manifolds and can be coupled across discretization levels. A key insight is that the previous state enters the light Schr\"odinger bridge only through a few features, so that the bridge only has to be fitted on the new manifold, regardless of the type of the previous one. We bound the bias of this particle filter and the gap between its approximations at consecutive levels, and show numerically, on spheres and on manifolds that change type, that this gap is much smaller than with standard multilevel particle filters, and that, with the cost of the fits included, a hybrid estimator that uses it at the fine levels reduces the cost at small tolerances.
Leticia Wüthrich, Mark Baillie, Susanne Dandl
Wed 18 Nov
stat.ME
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.89
- favorites near “Bayesian Semiparametric Model for Sequential Treatment Decisions with Informative Timing” 0.86
- interest Reinforcement learning, policy learning and off-policy evaluation 0.81
Abstract
Adjusting for prognostic baseline covariates in the analysis of randomized controlled trials is recommended to increase the precision of treatment effect estimates. Such adjustment, however, introduces a complication known as noncollapsibility: for effect measures such as the log-odds or log-hazard ratio, the estimated treatment effect becomes conditional instead of marginal. Methods that target a marginal estimand even after covariate adjustment have been developed, but typically rely on population-based inference, which assumes patients are randomly sampled from some larger population, rarely true in practice. Such methods have been shown to inflate type I error, particularly in small samples or when many covariates are adjusted for. Inference based on randomization requires no such assumption, as it relies on the randomization scheme, which is known exactly, yielding a test that is exact even in finite samples. We present a randomization-based test for covariate-adjusted marginal treatment effects, built on nonparanormal adjusted marginal inference, applicable to continuous, binary, ordinal, and time-to-event outcomes. In simulation studies, this test controls type I error where the standard Wald test does not. Applied to 20 historical phase III time-to-event randomized trials, the covariate-adjusted estimator achieves precision gains comparable to an existing competitor method.
Jun Zhang, Yongqiang Wu, Dan Pu, Wei Lan, Guanhao Feng
Tue 17 Nov
stat.ME
pdf html
- favorites near “Eight Centuries of Global Real Interest Rates, R-G, and the 'Suprasecular' Decline, 1311-2018” 0.78
- interest Causal inference and identification in econometrics 0.84
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.78
Abstract
We propose a Hierarchical Group Factor Model (HGFM) for jointly estimating global and group-specific factors with characteristic-driven portfolio weights. Both factor types are constructed from contemporaneous returns, producing endogeneity that we address by quasi-maximum likelihood estimation. Conditional on the interaction matrices, we separate the global and local coefficient components and normalize their factor representations. Under a fixed characteristic design and suitable regularity conditions, we establish local consistency of the coefficient estimators and asymptotic normality for fixed blocks of the normalized factor parameters as both panel dimensions diverge. We also propose an information criterion and establish its consistency for jointly selecting the global rank and the common local rank. In an industry-grouped application to S&P 100 constituents, the HGFM achieves higher explanatory power for realized excess returns and a higher portfolio Sharpe ratio than benchmark asset-pricing models.
Eliezer Pita Zambrano, Miguel Angel Luque-Fernandez
Fri 13 Nov
stat.ME
pdf html
- interest Online learning and statistical learning theory 0.88
- favorites near “Distributional Random Forests: Heterogeneity Adjustment and Multivariate Distributional Regression” 0.87
- your work near “Valid Inference with Imperfect Synthetic Data” 0.81
Abstract
The selection of an optimal predictive model is a pervasive challenge in biostatistics and epidemiology. Relying on a single a priori algorithm risks overfitting and reduced generalizability when applied to new settings. Super Learner is a flexible ensemble machine-learning method that mitigates these risks by combining multiple candidate algorithms into a single prediction model. Using cross-validation, it defines a weighted combination of learners that minimizes a specified loss function, ensuring performance asymptotically equal to or better than the best candidate algorithm. This tutorial provides a comprehensive, practical guide to implementing Super Learner in R for applied researchers. Using a simulated epidemiological dataset with a binary outcome, we demonstrate the step-by-step construction of the ensemble. We cover the foundational statistical concepts, basic implementation, evaluation through cross-validation, hyperparameter tuning, feature screening, and performance optimization for metrics such as the Area Under the ROC Curve (AUC). Additionally, we detail advanced usage, including computational parallelization and the integration of custom algorithms and screening procedures. The guide also includes a consolidated section on best practices for specifying SL, covering loss function selection, effective sample size determination, CV scheme definition, library curation, and discrete vs. ensemble SL decisions, plus a practical reference to the caret package for model preprocessing and comparison. This guide equips epidemiologists and clinical researchers with the tools to confidently incorporate robust machine learning techniques into their predictive modeling workflows.
Sheng Pan
Fri 13 Nov
stat.ME
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.86
- interest Causal inference and identification in econometrics 0.85
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.74
Abstract
In spatial risk attribution, continuous environmental exposures and unmeasured background factors often co-vary across broad geographic scales, making it difficult to separate true exposure effects from spatial confounding using finite disease counts. Existing remedies typically project out or penalize shared spatial trends, leaving unclear whether estimation instability stems from the overall capacity of the latent random field or from its directional alignment with exposure. By decomposing the Fisher-weighted smoothing operator of an additive Poisson point process model, we show that while effective degrees of freedom measure total smoothing capacity, the relative loss of Fisher information for the exposure coefficient is governed solely by a distinct directional quantity that we call penalized collinear overlap. Rotating the spatial kernel alters this information loss while holding total capacity fixed, implying that no single smoothing scale is universally optimal. Because real-world confounding combines smooth regional trends with sharp local jumps that escape any moderate basis dictionary, point identification remains fragile. We therefore develop a partial identification framework that bounds residual confounding bias using an observable benchmark covariate. Projecting a joint spatial confidence ellipse yields valid intervals even under weak benchmarks, while incorporating a negative control outcome sharpens the bounds and flags assumption violations. Simulation studies and an analysis of childhood respiratory mortality across England demonstrate the method.
Dragi Anevski, ElBatoul Manel Merai
Tue 17 Nov
math.STstat.TH
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.87
- interest Causal inference and identification in econometrics 0.82
- reading nonparametric, indirect, late, describes
Abstract
We study a graphical model (DAG) with time to event as a response variable. We use the Aalen additive hazard model for modelling the effect on the response variable of covariates, combined with a multiple linear regression for earlier layers in a DAG. This enables modelling covariates affecting the event process both directly and indirectly. The indirect component effect is represented by a product $\theta_1(t)\beta_2(t)$, where $\theta_1(t)$ describes the effect of the independent covariate on the dependent covariate and $\beta_2(t)$ is the effect of the dependent covariate on the final time to response variable, in the additive hazard model. We estimate $\beta_2(t)$ by the order-restricted estimator recently introduced in Anevski and Merai (arXiv:2606.23882v2), thereby enabling disease models with early onset or late onset covariate effects. Combining our previously obtained results with the DAG allows for modelling early onset or late onset covariate effects together with direct and indirect covariate effects, which should be of particular interest for studying causality for complex diseases. We present a nonparametric plug-in estimator of the indirect effect. We derive the limit distribution of the plug-in estimator of the indirect component, and show that it has $n^{1/3}$ pointwise asymptotic rate and a limit distribution which is proportional to the Chernoff distribution.
Daisuke Inoue, Dante Kalise
Mon 16 Nov
stat.MLmath.OC
pdf html
- reading grids, functions, kernels, nominal
- favorites near “Neural Operator: Learning Maps Between Function Spaces With Applications to PDEs” 0.81
- interest Computational methods for heterogeneous agent and dynamic economic models 0.78
Abstract
We propose mesh-free weak-form regression for learning interaction kernels from noisy particle trajectories. Our method uses smooth test functions to avoid differentiating noisy data and empirical particle averages to avoid density reconstruction and spatial grids. We then fit sparse combinations of candidate functions, evaluating interaction terms on sampled particle pairs to avoid all-pair quadratic cost. By transferring identifiability from the mean-field limit, we prove high-probability coefficient error bounds that separate finite-particle, time-discretization, observation-noise, and pair-sampling errors, without grid-resolution or density-estimation terms. We evaluate one-dimensional (1D) Newtonian-repulsion, 2D predator-prey, and 10D distance-localized experiments. The numerical results show trends consistent with the theoretical coefficient-error bounds. Comparisons of weak and forward-difference formulations with LASSO, Ridge, and OLS show that weak LASSO gives the smallest coefficient and drift errors in the three nominal experiments.
Shengbo Wang
Fri 13 Nov
math.OCcs.LGstat.ML
pdf html
- interest Computational methods for heterogeneous agent and dynamic economic models 0.81
- favorites near “Finding Regularized Competitive Equilibria of Heterogeneous Agent Macroeconomic Models via Reinforcement Learning” 0.79
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.80
Abstract
Q-learning has been empirically observed to exhibit substantial transient error. This transient remains uncaptured by the leading Gaussian limit because the central-limit normalization removes the smaller-scale displacement generated by the initial nonlinear evolution. On the other hand, worst-case finite-sample guarantees, while accounting for transient effects, can obscure the mechanisms responsible for the error on a particular problem instance. We develop a finer asymptotic analysis that gives an asymptotically exact characterization of this transient for a fixed instance. Specifically, for synchronous Q-learning in finite discounted Markov decision processes, we separate the last-iterate error into an optimal-policy temporal-difference (TD) component and a nonnegative displacement caused by maximization. The centered TD component has Gaussian fluctuations at scale $n^{-1/2}$, whereas the maximization transient is of order $n^{-1}$ and converges almost surely, after rescaling by $n$, to a random limit. This limit represents the propagated effect of maximization errors generated before the greedy policy stabilizes. We combine this characterization with an Edgeworth expansion of the TD component to identify the corrections to its Gaussian approximation, separating the effects of initialization, non-Gaussian TD fluctuations, and maximization. The results reveal how transient errors arise for individual Q-learning instances and how their effects persist below the central-limit scale.
Roberto Vila, Frederico Almeida
Tue 17 Nov
stat.ME
pdf html
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.79
- interest Causal inference and identification in econometrics 0.82
- favorites near “Labor Rationing” 0.77
Abstract
This paper introduces a rank-associated measure of population heterogeneity that quantifies how population differences are reflected in the relative positions of observations within a pooled sample. The measure is defined as a weighted average of Kolmogorov-Smirnov distances between conditional and unconditional distributions of order statistics, conditional on the population of origin at each rank. We establish its main properties, including boundedness, invariance under common strictly monotone transformations, and a characterization of its null value through independence between order statistics and population indicators. We derive bounds relating the measure to weighted pairwise distributional distances and population diversity across ranks, and examine the contributions of the minimum and maximum, obtaining a simple expression for two populations. Monte Carlo simulations with four populations assess its finite-sample behavior and asymptotic convergence. The empirical application reveals a shift in heterogeneity toward higher ranks in the germ-free group.
Baiyu Chen, Jun Fan, Zhou Yu
Tue 17 Nov
stat.ME
pdf html
- favorites near “Neural Operator: Learning Maps Between Function Spaces With Applications to PDEs” 0.83
- interest Kernels, operators and functional data 0.85
- reading functional, nonlinear, functional data, martingale
Abstract
We develop a modular framework for deep nonlinear Sufficient Dimension Reduction (SDR) of discretely observed functional predictors with Euclidean or metric-space-valued responses. A presmoothing encoder converts the functional input into finite-dimensional coordinates, and a neural network learns a sufficient representation from them. The framework separates the choice of encoder from that of the SDR criterion. We study kernel embedding and discrete functional principal component analysis, each combined with either a generalized martingale difference divergence network or a Fr\'echet cumulative covariance network. These combinations share objectives of the martingale difference divergence form. {Under input smoothness and general H\"older regularity of the target operator, we establish approximation and generalization bounds that jointly account for functional truncation, network approximation, statistical estimation, numerical integration, measurement noise, and encoder estimation. The complete theoretical analysis rigorously demonstrates how errors in the encoded coordinates propagate through the network to the deep nonlinear SDR estimator, and quantifies the final estimation error in relation to observation frequency and model complexity.} Simulations and an application to satellite-derived vegetation trajectories and crop-composition responses illustrate the statistical and computational advantages over RKHS-based nonlinear functional SDR methods.
Aukosh Jagannath, Varnan Sarangian
Tue 17 Nov
stat.MLcs.LGmath.PRmath.STstat.TH
pdf html
- interest Computational methods for heterogeneous agent and dynamic economic models 0.81
- reading laplacian, class algorithms, online, algorithms particular
- favorites near “Recent Developments in Machine Learning Methods for Stochastic Control and Games” 0.77
Abstract
We develop the high-dimensional effective dynamics for adaptive, momentum-based variants of Stochastic Gradient Descent (SGD). These effective dynamics are given by a Markov process. They key observation is that the non-Markovianity in the algorithm due to the exponential moving average and preconditioning can be treated using the method of averaging. Our approach applies to a broad class of algorithms including to Adaptive momentum estimation (ADAM), Momentum orthogonalized by Newton-Schultz (MUON), and its variants. It also applies to a broad class of learning tasks, such at tensor PCA, Gaussian projective models, teacher-student networks, and a multi-layer neural network trained on the XOR problem using a hybrid of MUON and ADAM as commonly done in practice. We demonstrate how to this technique can give precise, quantitative comparisons between algorithms. In particular, we find interesting phenomena, such as the cross-over phenomenon where online SGD has better generalization than ADAM early in training until the learning curves cross and ADAM dominates. Furthermore, we prove a version of the Edge of Stability phenomenon. In particular, we introduce an analogue of the infinity Laplacian that sharply characterizes the Edge of Stability.
Meenu Mohan, Siddhartha Chattopadhyay
Thu 19 Nov
econ.EMstat.AP
pdf html
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.82
- reading economic, wealth, economies, long-run
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.75
Abstract
This paper develops an Economic Progress Index (EPI), a composite measure built from Earth Observation (EO) and open geospatial data with unsupervised Principal Component Analysis (PCA) weighting. The index captures the observable productive and enabling capacity of local economies and the intensity with which economic space is used. Sub-national monitoring in developing economies is held back by sparse data: district GDP estimates arrive with multi-year lags, rely on price deflators and top-down allocation, and do not exist at the subdistrict (taluka/tehsil) scale where place-based policy is targeted. The EPI fuses multi-sensor satellite data, gridded population and OpenStreetMap road and amenity networks across 357 subdistricts of Maharashtra, India, over 46 quarters (2015 Q1 - 2026 Q2). Seasonal-Trend decomposition using LOESS (STL) separates long-run trends from monsoon and crop-cycle seasonality. A non-negative PCA gives sign-consistent, interpretable weights and rankings nearly identical to standard PCA ($\rho$ = 0.998). Across districts, the index explains 74% of the variation in the NFHS-5 wealth index, 52% in household consumption (HCES 2023-24) and about 78% in real district GDP levels (65% in logs), but does not track short-run output growth. Because it measures structural capacity rather than annual output, the index kept rising through the FY2020-21 COVID-19 contraction. Across Maharashtra's six revenue divisions, relative inequality fell by 44% even as the absolute gap between leading and lagging divisions widened by 32%; district-level $\beta$-convergence implies a half-life of about 14 years. The framework offers a scalable, deflator-free basis for structural monitoring in data-sparse economies.
Arya Farahi, Ritwik Vashistha
Fri 13 Nov
stat.MLcs.LGstat.ME
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.83
- interest Adaptive data collection and anytime-valid inference 0.86
- favorites near “Orthogonal Statistical Learning” 0.86
Abstract
Real-world machine learning systems often operate on imperfect measurements: sensors are noisy, annotations are produced by non-experts, and scientific observations are often indirect. In these settings, classic distribution comparison tools such as maximum mean discrepancy (MMD) can be systematically biased, matching artifacts of the measurement process rather than the underlying mechanisms of interest. We study a practical regime in which noisy observations are abundant, but only a small audited subset contains ground-truth measurements. We propose measurement-error-corrected MMD (MEC-MMD), an audit-based estimator that debiases kernel evaluations without requiring a noise model. MEC-MMD is exactly unbiased for the oracle MMD under simple random auditing, admits an explicit variance decomposition, and is consistent for the population MMD, achieving the standard $O_p(N^{-1/2})$ rate when the audit size scales with the dataset. We demonstrate the utility of distributional matching with MEC-MMD loss for both parameter estimation and unsupervised applications, showing that MEC-MMD consistently outperforms MMD and alternative methods. These results provide a principled foundation for robust distribution comparison, simulation-based inference, and learning in noisy real-world machine learning pipelines.
Jiayang Xu, Yusen Jia, Yangzhou Chen, Xin Chen
Wed 18 Nov
stat.ME
pdf html
- interest Semiparametric inference and debiased machine learning 0.90
- favorites near “Simplifying debiased inference via automatic differentiation and probabilistic programming” 0.86
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.83
Abstract
When the error is independent of the predictors, its distribution can carry information about a single-index direction beyond the conditional mean. We develop Single-Index Kernel Density Regression (SI-KDR): at each candidate direction, it refits the link and evaluates a leave-one-out kernel log-likelihood of the residuals. Under Gaussian design, the population score vanishes at the true line at every positive smoothing tuple. At a fixed positive-curvature tuple, the leading covariance in the sample expansion is the conditional-mean efficiency bound times $\Lambda/G$, where $G$ is the information gain from independence and $\Lambda$ is the variance inflation from smoothing. We derive $\Lambda$ by following each observation through link refitting, the dual role of its residual as evaluation point and kernel center, and training standardization. Case-weight covariance estimates guide tuning for point estimation. Tuning separately at each candidate gives an exact conditional $F$ test. When predictors outnumber observations, independent screening that retains the support preserves the covariance comparison and the exact test, and $G/\Lambda$ becomes the squared-noncentrality ratio of a retained-coordinate test relative to the conditional-mean oracle. In four non-Gaussian error distribution designs at $n=600$, covariance-trace tuning reduced direction risk by 56% to 74% relative to reference bandwidths, and score-inversion coverage was 93.8% to 96.6%.
Jianbin Tan, Anru R. Zhang
Tue 17 Nov
stat.MLcs.LG
pdf html
- reading american, discussion, american statistical, conditional distributions
- interest Language models as statistical objects 0.84
- favorites near “Deep Learning for Individual Heterogeneity” 0.80
Abstract
We discuss the Journal of the American Statistical Association (JASA) discussion paper "Deep Fr\'echet Regression" and explore an alternative perspective based on conditional flow matching. We propose learning covariate-dependent transport directly within the response space, avoiding separate manifold learning and reverse mapping. This perspective offers a potentially more flexible framework for incorporating geometric structure and modeling full conditional distributions.
Omer Haq
Fri 13 Nov
cs.LGcs.AIcs.CLstat.ML
pdf html
- favorites near “Sequential Monte Carlo Steering of Large Language Models using Probabilistic Programs” 0.89
- interest Language models as statistical objects 0.84
- reading state, language, evidence, language model
Abstract
Language models rely on finite token contexts, limiting access to earlier interactions. We introduce Recursive State-Augmented Language Models (Recursive SALM), a framework in which a persistent latent state evolves with each observation and generates adaptations of a pretrained language model. A modular read-transition-write architecture separates observation encoding, latent dynamics, and predictor adaptation. We develop deterministic and stochastic formulations. The experiments evaluate only deterministic SALM with frozen GPT-2 small and exclude earlier evidence from its decoder context. On a four-turn hypothesis-inference task, latent-gated LoRA reaches 89.1% accuracy, comparable to 88.0% for a trained full-context LoRA baseline, with earlier evidence available solely through a 64-dimensional state. Zeroing or shuffling the state substantially reduces accuracy. In a two-turn task with codewords unseen in training, several interfaces improve candidate ranking over chance, but rarely generate the exact codeword under the evaluated training schedule. These experiments show that persistent latent-conditioned adaptation can support sequential inference when earlier evidence is absent from the decoder context.
Peiyao Liu, Hongtao Zhu, Sizhe Zhang, Yang Su, Zhenyu Zhao, Nan Chen
Thu 19 Nov
stat.MEcs.LG
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.87
- reading causal, treatment, nonlinear, sequential
- interest Reinforcement learning, policy learning and off-policy evaluation 0.80
Abstract
Causal inference seeks to learn how outcomes change under alternative interventions, but network interactions complicate this task. Under network interference, outcomes depend on both own treatment and neighbors' treatments, making causal effects functions of neighborhood exposure rather than single treatment contrasts. Existing methods primarily target prespecified causal contrasts or their estimation precision, leaving global response-function recovery largely unexplored. We study sequential experimentation under a limited experimental budget, with the goal of recovering the full exposure-dependent population-average response function. To this end, we propose FANCI (Fill-Distance-Guided Active Learning for Network-Interference Causal Inference). For function learning, treatment-specific Gaussian processes flexibly model nonlinear conditional response surfaces. For experimental design, we formulate global response-function recovery as a coverage problem and propose a fill-distance criterion that guides new experiments toward poorly covered regions. To account for network constraints on attainable exposure patterns, we develop full-network and target-screened strategies for selecting feasible network--assignment pairs. Theoretically, we derive uniform error bounds linking fill distance to response-function estimation and characterize the approximation loss from target screening. Simulations and an online-game experiment show that FANCI improves recovery of nonlinear exposure-dependent effects and reveals treatment patterns obscured when network interference is ignored.
Yu Gui, Rina Foygel Barber, Cong Ma
Tue 17 Nov
math.STcs.LGstat.MLstat.TH
pdf html
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.81
- reading long-run, empirical, online, likely
- favorites near “Hierarchies of Relaxations for Online Prediction Problems with Evolving Constraints” 0.81
Abstract
Consider an online prediction setting, where at each time, we are given a new feature vector and asked to provide a prediction interval or prediction set that will likely contain the corresponding response. In this setting, if the data points are i.i.d. (or more generally, exchangeable), conformal prediction provides a mechanism for building valid prediction sets around the output of any trained model, offering coverage probability $\geq 1-\alpha$ at each time $t$. In practice, however, issues such as distribution drift, changepoints, or other challenges may cause violations of the i.i.d. assumption. In this adversarial setting, online conformal prediction methods can still offer guarantees on long-run empirical coverage. However, there is a trade-off between these two types of goals: enforcing tight constraints on long-run empirical coverage to protect against the adversarial setting can lead to unstable oscillations in the instantaneous training-conditional coverage in the i.i.d. setting. In this work, we seek best-of-both-worlds algorithms that aim to minimize cumulative error in instantaneous training-conditional coverage when in the i.i.d. setting, while also offering long-run empirical coverage guarantees in the adversarial setting. Specifically, consider a condition that requires long-run empirical coverage to deviate from the target level $1-\alpha$ by at most $\mathcal{O}(T^{-\beta})$, for some $\beta\in[0,1]$. We prove that this family of conditions can be described with a phase transition. For $\beta\in[0,\frac{1}{2}]$, it is possible to construct algorithms satisfying such a constraint in the adversarial setting, while offering the same instantaneous coverage performance as split conformal prediction if the data is in fact i.i.d. On the other hand, for $\beta\in(\frac{1}{2},1]$, this type of condition comes at a cost, and does not achieve best-of-both-worlds behavior.
Thomas Stark, Georg Köstenberger, Nina Dörnemann
Fri 13 Nov
math.STmath.PRstat.TH
pdf html
- interest Semiparametric inference and debiased machine learning 0.87
- favorites near “Eigendecompositions of Transfer Operators in Reproducing Kernel Hilbert Spaces” 0.84
- reading theory, tuned, attain, interpolation
Abstract
Many results in random matrix theory and high-dimensional statistics depend on weak convergence of spectral measures, which is usually established via pointwise convergence of Stieltjes transforms. This approach has recently been applied to signed measures. However, for signed measures, pointwise convergence of Stieltjes transforms alone no longer characterizes weak convergence. While this is unproblematic in many cases, in the high-dimensional regression setting, recently considered by Green and Romanov (2025), the underlying signed eigenvector-overlap measures can fail to converge weakly, despite pointwise convergence of their Stieltjes transforms. Using a novel approach to risk analysis, we show that weak convergence can be replaced by a simple convergence result for bilinear forms. This allows us to establish convergence of the prediction risk for estimators based on arbitrary continuous spectral filters, including (regularized) gradient descent and ridge regression, even when the underlying signed measures fail to converge weakly. Under additional regularity conditions, we construct an explicit finite signed measure yielding an integral representation of the asymptotic prediction risk. This representation provides necessary and sufficient conditions for minimum-norm interpolation and optimally tuned ridge regression to attain the smallest asymptotic prediction risk among estimators with continuous spectral filters. Moreover, we provide characterizations of weak and vague convergence of finite signed measures in terms of pointwise convergence of their Stieltjes transforms.
Elena Dal Torrione, Chan Park, Laura Forastiere
Mon 16 Nov
stat.MEecon.EM
pdf html
- reading treatment, policy, subject, population average
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.80
- interest Reinforcement learning, policy learning and off-policy evaluation 0.81
Abstract
Under heterogeneous interference, treating individuals with particular characteristics affects the population average outcome through their heterogeneous direct response to treatment and indirect effect on others, creating opportunities for efficient policy design. For instance, policymakers may minimize resource use by vaccinating individuals identified as superspreaders to achieve a target reduction in disease incidence. In such setting, we first study the problem of learning individualized stochastic treatment policies that minimize population average treatment probability subject to a prespecified outcome target. We propose an estimation approach based on a smooth surrogate for constraint violations and interference-aware inverse-probability-weighting or self-normalized estimators of the population average outcome, and establish finite-sample regret guarantees for both estimators. We also consider the complementary problem of maximizing the population average outcome subject to a constraint on average treatment probability. We apply the proposed methodology to a Colombian conditional cash transfer program aimed at improving school attendance. The application demonstrates that accounting for interference and individualizing treatment probabilities can reduce resource use relative to policies that ignore interference or impose homogeneous treatment probabilities.
Gaël Poux-Médard, Kaniav Kamary, Yohann De Castro, Vivian Viallon
Mon 16 Nov
stat.MLcs.LGq-bio.QMstat.ME
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.90
- interest Semiparametric inference and debiased machine learning 0.86
- reading rich, natural, structural, make
Abstract
Mediation analysis seeks to understand how an exposure influences an outcome through intermediate variables. However, as increasingly rich datasets become available, hundreds to thousands of potential mediators make these systems difficult to model and hard to interpret. Latent-variable approaches offer a natural way to accommodate these high-dimensional settings. However, existing latent mediation methods face limitations, including overparameterization, biased inference, or limited scalability. We develop LaMe, a joint linear-Gaussian framework that summarizes a high-dimensional exposure-mediator-outcome system through a small number of latent variables. We identify natural direct and indirect effects under an explicit assumption that the manifest measurements fully capture the latent mediators, and characterize conditions under which the latent representation effectively restricts the observed cross-covariance structure. Simulations assess recovery of observed association parameters, natural effects, and manifest features associated with latent mediation pathways across a range of challenging settings. Four real-world applications demonstrate competitive predictive performance with compact latent representations, including single-dimensional representations of biomarker and proteomic measurements. Together, these results establish LaMe as a practical framework for estimating and summarizing multivariate mediation systems under explicit structural and measurement assumptions.
Paul Grunenwald, Benoît Crabbé, Guillaume Wisniewski, Pirmin Lemberger
Fri 13 Nov
cs.CLstat.ML
pdf html
- favorites near “Using Imperfect Surrogates for Downstream Inference: Design-based Supervised Learning for Social Science Applications of Large Language Models” 0.89
- your work near “Valid Inference with Imperfect Synthetic Data” 0.82
- interest Language models as statistical objects 0.81
Abstract
In NLP, models trained via Empirical Risk Minimization (ERM) often exploit spurious correlations, such as dialectal markers, at the expense of robust linguistic features, thereby degrading their out-of-distribution (OOD) generalization. Invariant Risk Minimization (IRM) aims to suppress such shortcuts by identifying predictive relationships that remain stable across environments, but its practical utility for NLP remains debated. We use a causality-driven framework to evaluate IRM under three shortcut-generating mechanisms: label-generated features, confounding, and selection bias. Our results show that IRM successfully rejects shortcuts across the evaluated causal and anti-causal structures, including under limited environmental variation and increasing label noise. We then extend our analysis beyond controlled synthetic artifacts to two less constrained settings: gender-profession associations in Bias in Bios and semantically loaded actor-name shortcuts in ActorCorr. These experiments show that invariant learning can mitigate more realistic textual shortcuts, while semantically meaningful features require careful calibration of the invariance penalty.
Geng Zhang, James C. Spall
Tue 17 Nov
stat.AP
pdf html
- reading quadrature, derivatives, gives, matrices
- favorites near “Monte Carlo Geometry Processing” 0.82
- interest Semiparametric inference and debiased machine learning 0.83
Abstract
The log-determinant of a positive-definite Hessian matrix is used in uncertainty assessment, Bayesian model comparison, and experimental design. Computing this quantity can be difficult when function values are available but derivatives are not. We combine simultaneous perturbation stochastic approximation with stochastic Lanczos quadrature (SPSA-SLQ) to estimate the log-determinant without forming the Hessian. Possibly noisy function values suffice for fixed-data and expected Hessians, including Fisher information matrices. We derive a local bound on log-determinant error due to Hessian-vector product approximation, quadrature, and averaging random probes. Three numerical studies examine accuracy and computational cost. In a 3002-parameter example, SPSA-SLQ gives a relative root-mean-square error below 1% and uses 65% less peak GPU memory than a basic full-matrix estimator. A sensor-placement study also compares improved full-matrix methods and evaluates layout selection.
Xiaochen Zhang, Wenjing Zheng, Xingye Kang, Vamsi Manthena
Mon 16 Nov
stat.APecon.EM
pdf html
- reading economy, treatment, little, reach
- interest Adaptive data collection and anytime-valid inference 0.84
- your work near “Efficient Online Estimation of Causal Effects by Deciding What to Observe” 0.81
Abstract
Roblox spans many product surfaces-discovery, the virtual economy, creator tools-that differ sharply in reach and in how much the affected population contributes to any sitewide metric. A randomized experiment shows whether a treatment works for the users in its sample, but this experiment lift is not comparable across surfaces, because surfaces differ in reach and in their contribution to sitewide metrics. The quantity a strategic decision needs is a counterfactual-how much an aggregated sitewide metric (e.g., total platform play time) would move if the treatment reached 100% of its target population, which may be only part of the sitewide population, during the measurement period. We call this sitewide impact. Its scaling factor, which we call the coverage factor, combines the reach and the target population's contribution to the sitewide metric under no treatment. Hence a low-reach surface whose users carry much of the metric could outrank a high-reach surface whose users carry little. We develop two estimators under this framing: one for difference-in-means metrics and one for variant-level ratio metrics (a metric defined as a ratio of two population aggregates). Empirically, experiment lift is a poor proxy for sitewide impact-across 139 experiments that are statistically significant on time spent, the Spearman rank correlation between the two is 0.20, and 5 of the 10 largest experiment lifts are among the 10 largest sitewide impacts. We validate the estimator with A/A tests and against a long-term holdout, and apply it to cross-domain prioritization and attributing year-over-year growth to shipped innovations.
Nicholas Tenkorang, Emmanuella Duah, Bismark Nyarko
Tue 17 Nov
stat.ME
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.86
- reading semiparametric, adjustment, treatment, nuisance
- favorites near “Bayesian Semiparametric Model for Sequential Treatment Decisions with Informative Timing” 0.85
Abstract
Overlap weighting addresses limited treatment overlap by targeting individuals for whom treatment assignment is uncertain. With right censored outcomes, however, adequate treatment overlap does not ensure adequate follow up. We show that the overlap weighted restricted mean survival time difference can have an infinite semiparametric efficiency bound even under randomized treatment and pointwise censoring positivity. We propose censoring adjusted overlap (CO) weighting, that accounts jointly for treatment assignment and the probability of remaining under observation through the restriction horizon. CO approaches ordinary overlap weighting as the strength of the follow up adjustment decreases and progressively downweights covariate profiles with weak follow up. We derive the efficient influence function and show that it is uniformly bounded without requiring censoring survival probabilities to be uniformly bounded away from zero. A cross fitted one step estimator is asymptotically normal with consistently estimated variance under follow up weighted nuisance rate conditions. Simulations show little difference from overlap weighting when follow up is adequate, but substantially greater stability as follow up deteriorates. An analysis of the Rotterdam breast cancer cohort illustrates both the protection against weak follow up and the potential cost in precision.
Peng Zeng, Hanwen Huang
Fri 13 Nov
stat.ME
pdf html
- interest Semiparametric inference and debiased machine learning 0.88
- favorites near “Advances in Nowcasting Economic Activity: Secular Trends, Large Shocks and New Data” 0.80
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.79
Abstract
Li (1991) introduced sliced inverse regression (SIR) as a dimension reduction method and established its theoretical properties under the assumption that the number of predictors $p$ remains fixed as the sample size $n$ tends to infinity. In this paper, we revisit the theoretical properties of SIR in the high-dimensional regime where $p/n \to \gamma \in [0,1)$, assuming normally distributed predictors and a fixed number of slices. We show that the sample SIR eigenvalues exhibit an upward bias when $\gamma>0$, that their fluctuations are governed by the Gaussian orthogonal ensemble, and that the signal eigenvalues are asymptotically normal after an appropriate bias correction. We demonstrate that Li's chi-squared test requires a scaling adjustment to achieve the nominal asymptotic size when $\gamma>0$. We further propose two additional tests that achieve the nominal asymptotic size: an exact test based on Li's test statistic and a largest-eigenvalue test. In addition, we derive closed-form limits for the accuracy of the estimated SIR directions and show that their first-order expansions in $\gamma$ agree with Li's approximation. Finally, we establish the limiting distribution of the estimated sufficient predictors for a new observation and show that, when $\gamma>0$, they contain an additional independent Gaussian noise component. Simulation studies corroborate our theoretical findings.
Vignesh Sella, Anirban Chaudhuri, Karen Willcox
Mon 16 Nov
stat.MLcs.CEcs.LGcs.NAmath.NA
pdf html
- favorites near “Neural Operator: Learning Maps Between Function Spaces With Applications to PDEs” 0.85
- interest Computational methods for heterogeneous agent and dynamic economic models 0.80
- reading operators, operator, dynamical, nonlinear
Abstract
This paper presents a multifidelity (MF) formulation of Operator Inference (OpInf) for learning non-intrusive, physics-based reduced-order models (ROMs) of nonlinear dynamical systems. OpInf learns reduced operators from projections of snapshot data by solving a least squares regression problem. However, generating high-fidelity snapshots can be computationally expensive, limiting parameter coverage and the temporal extent of training trajectories for many aerospace applications, which can lead to inaccurate or unstable ROM predictions. This work proposes MF-OpInf, which combines limited high-fidelity snapshots with more plentiful low-fidelity snapshots through a control-variates-inspired MF least squares formulation to learn a single set of reduced operators across multiple trajectories. Low-fidelity trajectories provide additional parameter coverage and extend beyond the high-fidelity training time horizon. The low-fidelity snapshots are also used to construct an improved reduced basis through weighted proper orthogonal decomposition. We demonstrate MF-OpInf on a nonlinear reaction-diffusion problem with parametrized sinusoidal forcing and a two-dimensional incompressible flow past a cylinder with a parametrized inlet pulse. In these experiments, under an equal training simulation budget, MF-OpInf reduces the mean relative prediction error at unseen parameters by up to $87\%$ for the reaction-diffusion problem and $28\%$ for the cylinder flow problem over the test time regime, which extends beyond the high-fidelity training time horizon. These results demonstrate that MF-OpInf can improve temporal extrapolation and prediction at unseen parameters without increasing the training simulation budget.
Nilava Metya, Oscar Leong, Yong Sheng Soh
Fri 13 Nov
math.OCmath.MGstat.ML
pdf html
- favorites near “Neural Operator: Learning Maps Between Function Spaces With Applications to PDEs” 0.84
- reading questions, ill-posed, ill-posed inverse, describing
- interest Kernels, operators and functional data 0.83
Abstract
We consider the question of describing gradient flows over the collection of convex bodies. This is motivated by foundational questions arising in inverse problems. Suppose we wish to solve an ill-posed inverse problem using regularization techniques. Can we learn an optimal choice of a convex regularizer directly from prior data? Given a numerical algorithm that computes such an optimal regularizer, is there a principled manner for describing the evolution of its iterates? The objective of this paper is to lay the foundation towards answering these questions. To this end, we consider a criterion of an optimal regularizer based on seeking a convex body that minimizes the expected gauge evaluation with respect to a prior data distribution subject to a volume constraint. The formulation is initially nonconvex, but admits hidden convexity by passing to its polar. We characterize the Karush-Kuhn-Tucker (KKT) conditions that certify global optimality. We also analyze the $L^2$-gradient flow as well as a minimizing movement scheme with the Hausdorff penalty. Our results show that the objective gap decays exponentially along the gradient flow and we obtain dimension-dependent rates for both the objective gap and distance to the optimal body for the Hausdorff scheme. Finally, we apply the KKT conditions to identify explicit optimal convex bodies. For data where the optimal nonconvex regularizer is the $\ell_p$-quasinorm, $p \in (0,1)$, the optimal convex regularizer is the volume-normalized $\ell_1$-ball. For data described by paired basis atoms, the optimal convex regularizer is the polar of a shifted zonotope.
Roberto I. Oliveira, Zoraida F. Rico, Philip Thompson
Fri 13 Nov
math.OCmath.STstat.TH
pdf html
- interest Adaptive data collection and anytime-valid inference 0.84
- your work near “Valid Inference with Imperfect Synthetic Data” 0.80
- favorites near “Orthogonal Statistical Learning” 0.87
Abstract
We study efficient robust estimation of a high-dimensional mean under strong contamination when the clean distribution has only $p\ge2$ moments, bounded by $\nu_p$. We give a unified proof of stability for every $p\ge2$ with the optimal dimension-free sub-Gaussian rate, linking the reweighting, trimming and truncation techniques of the outlier and heavy-tailed robust literatures through concentration of counting and truncation processes. Building on stability and spectral signatures, we quantify the optimization landscape over sample weights that remove at most a fraction $\varepsilon_{n,k}=k/n$ of the mass, where the trimming parameter $k$ slightly exceeds the number of outliers. An unobserved convex spectral loss has inexact quadratic growth, with curvature $\frac{1-\varepsilon_{n,k}}{\varepsilon_{n,k}}$, relative to a mean-induced Bregman divergence, and the observable spectral loss supplies a dimension-free, inexact, relatively weakly star-convex subgradient oracle for it. This yields a Lyapunov analysis of mirror descent reweighting for $\varepsilon_{n,k}<\frac13$ that covers Hedge, the projected subgradient method, a new mean-induced mirror descent, and a centered multiplicative weights update. Each method uses one top-eigenvector computation per step and attains the optimal sub-Gaussian rate without knowledge of $(p,\nu_p)$. Mean-induced mirror descent further achieves faster optimization rates and last-iterate guarantees, at the cost of a more expensive projection step in higher dimensions.
Christopher L. Hammond
Tue 17 Nov
stat.MLcs.LG
pdf html
- favorites near “A U-turn on Double Descent: Rethinking Parameter Counting in Statistical Learning” 0.88
- interest Kernels, operators and functional data 0.83
- reading nonparametric, curved, net, empirical
Abstract
We study when compositional structure yields dimension reduction in supervised learning. We frame learning as empirical risk minimization with a function-space prior: a regularization functional selects a subset of functions with favorable approximation and metric entropy. The models are compositions $f(x)=g(Ax+\Delta h(x))$, where $A\in\mathbb{R}^{k\times d}$ is a linear base map, $\Delta h$ is a nonlinear correction, and $g$ is an outer function; that is, $f$ is an arbitrary smooth function of a $k$-dimensional representation of the input that is linear up to a correction. We prove covering-number bounds showing that the statistical complexity of this class interpolates between an intrinsic regime, parametric in the $kd$ entries of the linear map and nonparametric only in the representation dimension $k$, and an ambient $d$-dimensional nonparametric rate; the interpolation is controlled by the correction size $\varepsilon=\|\Delta h\|$. We then connect $\varepsilon$ to data geometry. If the data lie in a small tube around a $k$-dimensional manifold and the regression target depends on the input only through its nearest point on that manifold, then locally the target factorizes exactly as $g(Ax+\Delta h(x))$ with $A$ a tangent chart and a correction whose Sobolev norm scales linearly with curvature at the localization scale. Patching such local models over an atlas gives a finite-sample excess-risk bound for a least-squares estimator over a finite net of local models, whose ambient term vanishes with curvature. For a one-dimensional model problem we also prove a converse: on curved data, every affine compositional predictor faces an error floor unless its correction budget scales with curvature, while the matching curved model class attains the intrinsic rate. The correction size $\varepsilon$ is thus the effective-dimension parameter linking curvature to statistical complexity.
Dominik Ammon, Julian Heinz
Thu 19 Nov
econ.EM
pdf html
- favorites near “Eight Centuries of Global Real Interest Rates, R-G, and the 'Suprasecular' Decline, 1311-2018” 0.78
- interest Macroeconometrics and time series 0.83
- reading dynamic, forecasting, specifications, out-of-sample
Abstract
We study how commonalities in monthly realized stock volatilities can be exploited for high-dimensional covariance forecasting and portfolio selection. We compare latent factor models with ARMA models that pool dynamic parameters within pre-specified groups of stocks. We show that pooled ARMA models admit an exact dynamic factor representation with group-specific dynamics and stock-specific loadings. The implied static factor representation involves group-specific factors that need not be pervasive. Combining volatility forecasts with realized dynamic conditional correlations and nonlinear shrinkage, we evaluate out-of-sample minimum variance portfolios of S&P 500 constituents. Pooling-based approaches consistently outperform the alternatives across portfolio constraints and model specifications.
Rida Benhaddou
Thu 19 Nov
math.STstat.MEstat.TH
pdf html
- reading wavelets, regular, wavelet, rates
- interest Semiparametric inference and debiased machine learning 0.85
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.81
Abstract
Simultaneous deconvolution model is investigated when the design points are random and follow bounded densities and errors are either $i.i.d.$ or long-memory dependent. We propose wavelet hard-thresholding estimators, and show that they achieve asymptotically near-optimal convergence rates for the $L^2$-risk when the response function belongs to Besov space under both $i.i.d.$ errors and homoskedastic, long-memory errors. However, under heteroskedastic, long-memory errors the estimator is not optimal. For the $i.i.d.$ errors, the convergence rates are {identical} to those found in simultaneous deconvolution under regular design {and $i.i.d.$ errors}. However, for long-memory dependent errors, the convergence rates are new {and completely different from those in simultaneous deconvolution under regular design and long-memory errors}.
Kosuke Morikawa, Yutaka Kano, Shonosuke Sugasawa
Fri 13 Nov
stat.MEmath.STstat.TH
pdf html
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.83
- interest Adaptive data collection and anytime-valid inference 0.85
- reading semiparametric, doubly robust, doubly, procedures
Abstract
The missing-at-random assumption underlies most analyses of incomplete data. Under nonmonotone missingness, it is especially stringent, imposing pattern-specific restrictions with different conditioning sets. A strictly weaker normalization condition is known to justify maximizing the observed-data likelihood. We show that every law satisfying this condition has a unique missing-at-random counterpart sharing its observed-data and complete-data distributions. Observed-data procedures developed under missing at random for identified functionals, including doubly robust estimators, therefore remain valid, with the same semiparametric efficiency bound. We then prove that the condition holds if and only if filling in the missing values from their conditional distribution under the true full model preserves the complete-data distribution, hence the name completion-invariant missingness. We use a characteristic kernel to compare the completed distribution with the fitted model and assess the joint compatibility of the specified full model and completion-invariant missingness. We derive the asymptotic null distribution with estimated parameters and establish bootstrap and multiplier calibration. Since a parametric full model alone need not identify completion-invariant missingness, we give identification conditions for Gaussian, Student $t$ and finite mixture families. Simulations assess sensitivity to departures from each assumption, and applications to clinical laboratory measurements and childhood growth illustrate the diagnostic.
Brandon Amos
Mon 16 Nov
cs.LGcs.NAmath.NAstat.ML
pdf html
- reading affect, computations, nearly, issues
- interest Language models as statistical objects 0.82
- your work near “Valid Inference with Imperfect Synthetic Data” 0.78
Abstract
Barron et al. (2026) recently identified numerical issues in the softmax and soft cross-entropy loss. We show that similar issues are present in standard divergence operations between categorical distributions. This includes the forward and reverse Kullback-Leibler (KL) divergence and generalized Jensen-Shannon divergence (JSD), as well as Monte Carlo KL estimates and entropy. When the distributions are nearly equal, rounding and cancellation issues affect the forward and gradient computations. These issues can affect settings using divergences in optimization objectives, such as knowledge distillation, reinforcement learning, and proximal methods. We propose numerically stable reformulations and provide example implementations for each setting, validated by property-based tests.
Wenjing Meng, Jeffrey E. Max, Florin Vaida
Tue 17 Nov
stat.ME
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.84
- reading children, treatment, log, nominal
- interest Causal inference and identification in econometrics 0.82
Abstract
Cohort studies and surveys often record how much time each subject spent under each of two or several treatment conditions and whether an event occurred, but not the condition under which it occurred. We develop two models for such data, for binary and count outcomes. Both start with events under each treatment condition as independent Poisson counts, and estimate rate ratios between treatment and control conditions. In the binary model, only an indicator of any event is observed, and the event probability under each condition follows a complementary log-log model with a log-exposure offset; development is done for general link functions, including log and logit. The interpretation of the treatment effect changes accordingly. In the count model, the number of events combined over all treatment conditions is observed; overdispersion is handled by Pearson scaling. Parameters are estimated either by the expectation-maximization algorithm or by direct maximization of the observed-data likelihood with Fisher scoring; standard errors are available via observed or expected information. When total exposure is common to all subjects, a relative slope summarizes the treatment effect and, unlike the rate ratio, does not depend on the assumed total exposure. In simulations, interval coverage was near nominal when enough events occurred during each treatment condition. The models are implemented in the R package maskcave. The work was motivated by the Adolescent Brain Cognitive Development study of 10,948 children. Assuming 16 waking hours per day, the adjusted sport-to-non-sport rate ratio for mild traumatic brain injury was 8.78 (95% confidence interval 4.47-17.27). We applied the count model to concussions during sports reported for 6,379 children at the Year 5 visit: the concussion rate per hour of football was 2.46 (1.37-4.43) times that of the other reported sports.
Hamed Karimi, Julie Nutini, Mark Schmidt
Mon 16 Nov
math.OCcs.LGstat.ML
pdf html
- reading hindsight, year, main, larger
- favorites near “Mean field games and applications” 0.84
- interest Computational methods for heterogeneous agent and dynamic economic models 0.78
Abstract
In 2016, we wrote a paper introducing the phrase "Polyak-Lojasiewicz (PL) condition" for an old idea, unsure of how readers would react. The paper ultimately had a larger impact than we expected, and we are surprised and honoured to receive the 2026 PKDD ECML Test of Time Award. With 10 years of hindsight, this note offers an informal and candid retrospective on the PL condition. We take a look back at the origins of the paper, give an AI-assisted refined version of the main theorem, review criticisms of the work, and discuss a variety of related ideas and extensions.
Xuran Meng, Yi Li
Fri 13 Nov
math.STstat.TH
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.83
- interest Adaptive data collection and anytime-valid inference 0.84
- reading semiparametric, large, theory, gains
Abstract
Transfer learning can improve precision by borrowing information from external sources, but how much improvement is possible while maintaining valid inference? We address this by studying honest confidence intervals for a target parameter using multiple sources with heterogeneous prespecified discrepancy bounds. Under local information and admissibility conditions, we establish a universal lower bound on interval length for procedures maintaining uniform coverage. A scalar information equation characterizes the bound, revealing how precision depends jointly on source sample sizes and discrepancies. Large source samples need not yield greater precision gains when their discrepancies are large. When discrepancies are unrestricted, honest intervals cannot improve upon the target-only rate, even when all sources are unbiased. For the Gaussian location model, we calculate optimal source weights and construct honest intervals attaining the lower bound order, establishing sharpness. We extend the lower bound theory to semiparametric models through parametric submodels. For the Cox proportional hazards model, we verify the information conditions under which the lower bound applies, including under right censoring.
Zhiguo Li
Fri 13 Nov
stat.ME
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.85
- favorites near “Bayesian Semiparametric Model for Sequential Treatment Decisions with Informative Timing” 0.87
- interest Reinforcement learning, policy learning and off-policy evaluation 0.80
Abstract
Win statistics provide clinically interpretable measures for comparing treatment groups with prioritized composite outcomes, but efficient covariate-adjustment methods for win statistics with right censored time-to-event outcomes remain limited. We develop a semiparametric framework for marginal win probability and win odds with prioritized time-to-event outcomes, including semi-competing risks, while also accommodating scalar survival outcomes as a special case. The estimand compares independent draws from the marginal distributions of the potential outcomes under treatment and control, which preserves a marginal causal interpretation. We derive the efficient influence function under conditional independent censoring, allowing censoring to depend on baseline covariates and predictable event history. The resulting cross-fitted one-step estimator incorporates covariate adjustment and censoring martingale augmentation, allows flexible estimation of nuisance parameters, and attains the semiparametric efficiency bound under suitable conditions on consistency and convergence rate of nuisance parameters. Simulation studies show that the estimator has small bias, accurate standard error estimation and confidence interval coverage, and substantial efficiency gain relative to the inverse-probability-of-censoring weighted (IPCW) estimator. An application to a randomized breast cancer trial comparing tamoxifen combined with radiotherapy against tamoxifen alone demonstrates improved precision, with an approximately 25\% reduction in estimated variance at five years.
Yimin Zhong, Jiaxin Zhou
Fri 13 Nov
stat.MLcs.LGmath.OC
pdf html
- reading functions, interpolation, stochastic gradient, multiscale
- favorites near “Recent Developments in Machine Learning Methods for Stochastic Control and Games” 0.78
- interest Computational methods for heterogeneous agent and dynamic economic models 0.78
Abstract
In this work, we study last-iterate convergence bounds for stochastic gradient descent (SGD) on convex and $\beta$-smooth loss functions and improve the bound from $\cO((\log T)^3/T)$ to $\cO((\log T)^3/(T\log\log T))$ in the interpolation regime. The improvement comes from applying a horizon-dependent multiscale estimate of gradients to the comparison of trajectories. The bound applies to both sampling with replacement and sampling without replacement within one pass.
Haoge Chang, Kota Saito, Mingxi Zhu
Tue 17 Nov
econ.EMecon.TH
pdf html
- interest Reinforcement learning, policy learning and off-policy evaluation 0.83
- your work near “Valid Inference with Imperfect Synthetic Data” 0.80
- reading choice, inequalities, available, computational algorithms
Abstract
This paper develops computational algorithms for testing discrete-choice random utility models (RUM). The set of RUM-consistent choice probabilities forms a polytope, and existing computational methods rely on its vertex representation. We exploit alternative half-space (Block--Marschak inequalities) and network-flow representations to develop two methods based on the alternating direction method of multipliers and a flow-based column-generation algorithm. Compared with current vertex-based methods, our algorithms are better suited to GPU implementation. We establish convergence guarantees and provide optimality certificates for all three algorithms. The resulting algorithms achieve $100$--$600\times$ speedups over existing methods. The solver based on our methods is publicly available on Github at \href{https://github.com/mingxiz/d-rum-testing-gpu}{\texttt{d-rum-testing-gpu}}.
Otto Tabell, Jani Nykänen, Santtu Tikka, Juha Karvanen
Mon 16 Nov
stat.MLcs.LGstat.ME
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.86
- interest Causal inference and identification in econometrics 0.81
- favorites near “Bayesian Workflow” 0.87
Abstract
Clustering variables in a graph may improve visual clarity and simplify causal inference. Harnessing the full potential of clustering requires an automated method to detect useful clusters in a graph. We introduce a novel class of clusters called conduits, and present a sound, complete and efficient algorithm for detecting all conduits in the graph. In addition to algorithmic detectability, conduits appear naturally in causal graphs stemming from real-world applications, enabling substantial size reduction. Furthermore, the structure of conduits enables the application of known identifiability results regarding clustering. We demonstrate these properties in a simulation study with real-world graphs.
Shunxing Yan, Fang Yao
Thu 19 Nov
math.STcs.LGstat.MLstat.TH
pdf html
- interest Semiparametric inference and debiased machine learning 0.87
- your work near “Efficient Online Estimation of Causal Effects by Deciding What to Observe” 0.81
- favorites near “Eigendecompositions of Transfer Operators in Reproducing Kernel Hilbert Spaces” 0.82
Abstract
Functional principal component analysis (FPCA) plays a central role as a dimension-reduction and representation tool in functional data analysis, but its minimax theory for diverging spectral indices under noisy discrete observation has remained open for over two decades. To fill this gap, we consider a sieve-based FPCA framework that pools discrete measurements across subjects and accommodates sampling frequencies ranging from sparse to dense under random designs. We begin by establishing new minimax rates for covariance estimation over general bivariate function classes beyond tensor-product low-complexity structures, thereby identifying the intrinsically two-dimensional optimal rate. Our central theoretical contribution is to establish minimax optimal rates for both individually labelled eigenfunctions with diverging indices and leading principal subspaces of increasing dimensions, under explicit growth and regularity conditions. To establish these results, we develop a general framework for sharp minimax analysis of FPCA under discrete observation, separating spectral approximation from stochastic perturbations and exploiting their distinct structures, thereby offering a new approach to a broad class of nonparametric spectral estimation problems. For both spectral targets, the optimal rates reveal two distinct statistical mechanisms, spectral identification and nonparametric reconstruction, whose interplay gives rise to either four or two sampling regimes. This yields, for the first time, a precise characterization of the impact of discrete sampling across the full range from sparse to dense designs. Finally, as a downstream application, we analyze FPCA-based functional linear regression and show that the FPCA-based estimator attains the minimax-optimal rates.
Hirofumi Shiba, Luke Hardcastle, Kengo Kamatani
Fri 13 Nov
stat.COstat.ME
pdf html
- reading manifold, versus, state, processes
- favorites near “Eigendecompositions of Transfer Operators in Reproducing Kernel Hilbert Spaces” 0.83
- interest Bayesian computation and structural estimation 0.83
Abstract
Assigning positive prior probability to lower-dimensional submodels enriches the range of uncertainty that Bayesian models can represent. The resulting posterior distributions combine continuous parameter uncertainty with discrete model uncertainty. Efficient movement between these submodels, however, can be difficult to design. We propose the sticky manifold construction as a principled, modular framework for combining piecewise deterministic Markov processes (PDMPs) across different submodels. The integration is rejection-free, requires no additional likelihood evaluations, and retains the non-reversible and ballistic nature of the component PDMP dynamics. We prove that the resulting process leaves the intended mixed-dimensional target invariant. We also develop a manifold PDMP algorithm based on coordinate charts to provide the within-submodel dynamics and a blow-up procedure that resolves singular sets into codimension-one sticky boundaries. This geometric construction makes the remaining design choices explicit; component dynamics and unsticking kernels can be chosen freely, subject to easily verifiable velocity compatibility conditions. Finally, we demonstrate the sticky manifold algorithms in Bayesian model averaging examples, propagating uncertainty over pure versus mixed states in quantum state tomography, rank in Bayesian principal component analysis, and parametric families in survival extrapolation.
Jinan Pan, Timo Schorlepp
Tue 17 Nov
stat.COcond-mat.stat-mechmath.OCmath.PR
pdf html
- reading equations, large, differential equations, riccati
- interest Computational methods for heterogeneous agent and dynamic economic models 0.78
- favorites near “Bernstein-von Mises theorems for time evolution equations” 0.87
Abstract
We extend the optimal control formulation of sample path large deviation theory, which has proven effective for numerical computation of instantons and sharp probability estimates in stochastic differential equations with Gaussian noise, to systems with light-tailed L\'evy noise. In the scaling limit of small jumps and high intensity, such systems satisfy a large deviation principle of Freidlin-Wentzell type, with a running cost given by the Legendre transform of the noise cumulant generating function. We propose to numerically minimize this cost over noise realizations subject to the system dynamics, which yields the instanton without having to solve Hamilton's equations, and naturally handles degenerate and multiplicative noise. When the cumulant generating function has a bounded domain, the cost grows only linearly for large noise, and instantons can contain jumps. We analyze when jumps are not only admissible but optimal, depending on the subleading behavior of the cost and on the drift, and show how jumps can be resolved numerically through a time reparameterization that freezes physical time. For continuous instantons, we derive Riccati and Fredholm determinant expressions for the leading-order prefactor, and we discuss the modifications needed when the instanton jumps. We illustrate and validate our methods against Monte Carlo simulations in several examples: a two-dimensional nonlinear model, an active run-and-tumble particle with non-Gaussian tumbling noise, and the Ghil-Sellers energy balance climate model with L\'evy-perturbed solar forcing.
Yantao Wu, Mauro Maggioni
Tue 17 Nov
stat.MLcs.LGmath.STstat.TH
pdf html
- interest Kernels, operators and functional data 0.86
- favorites near “Eigendecompositions of Transfer Operators in Reproducing Kernel Hilbert Spaces” 0.84
- reading nonparametric, manifold, dimensionality, multiscale
Abstract
We study vector-valued regression with predictors filling a tube around an unknown $m$-dimensional manifold $\mathcal{M}$ in $\mathbb{R}^D$. In the Manifold-Index Model, the mean response is $F=f\circ\Pi_{\mathcal{M}}$, where $\Pi_{\mathcal{M}}$ is the nearest-point projection onto $\mathcal{M}$ and $f:\mathcal{M}\to\mathbb{R}^p$ is a possibly non-injective link function. Multiscale ANisotropic lOcal Regression (MANOR) estimates tangent planes by multiscale PCA of predictors with similar responses, then refines them by pooling outer products of local slopes. The neighborhoods narrow tangentially while retaining complete normal fibers. We assume a Lipschitz link differential and bounded response noise. For $p\ge m$, under our geometric and sampling conditions, $n$ independent observations give mean tangent-space error of order $n^{-2/(m+4)}$ and integrated squared prediction error at the $m$-dimensional nonparametric rate $n^{-4/(m+4)}$, up to logarithmic factors. In a scaling regime with fixed intrinsic dimension, model and scale parameters, and ratio of manifold volume to the $m$-th power of its reach, the required sample size grows polynomially in $D+p$, and computation is polynomial in $n$, $D$ and $p$. Numerical experiments on curves with non-monotone links and surfaces with vector-valued responses compare observed convergence rates with the theoretical bounds; a Langevin dynamics example tests prediction of hitting probabilities under an approximate manifold-index model.
Chencheng Fang
Tue 17 Nov
stat.ME
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.81
- reading confidence sequences, series, control, finance
- interest Adaptive data collection and anytime-valid inference 0.81
Abstract
Sequential prediction on streaming time series is essential for continuous monitoring and risk control in applications such as finance, energy, and healthcare systems, where prediction intervals must remain valid at arbitrary stopping times and adapt to potential distributional shifts. Conventional conformal prediction provides model-agnostic uncertainty quantification but relies on exchangeability, which is violated under distributional shifts. Adaptive Conformal Inference (ACI) relaxes this assumption by updating the miscoverage level in response to the realized prediction errors, but it only guarantees that the realized average miscoverage rate along a single trajectory approaches the target level in the long run. Risk-sensitive monitoring instead requires controlling the true miscoverage rate, i.e., the conditional probability that a prediction set misses, which measures the reliability of the set itself rather than the misses that happened to be realized. We show that, without assumptions on the distributional shifts, it is impossible to have such control for every single prediction set at arbitrary stopping times. We therefore target the true running and rolling average miscoverage rates. Our methods combine ACI with anytime valid confidence sequences. With high probability, the true running average miscoverage rate stays below the target level uniformly over time, and so does the true rolling average for any window rule. The efficiency loss for the rolling average vanishes for windows that grow sufficiently fast, but not for fixed windows.
Tania Roa, Meryem Slaoui, Soledad Torres
Fri 13 Nov
math.STstat.MEstat.TH
pdf html
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.81
- interest Macroeconometrics and time series 0.84
- favorites near “Priors for the Long Run” 0.82
Abstract
Nonparametric regression conventionally assumes equispaced observations, a premise often violated by empirical data with irregular sampling and long-memory noise. This study establishes the asymptotic mean square ($L^2$) consistency of a weighted Nadaraya-Watson (wN-W) estimator for stochastic processes sampled at random times and perturbed by fractional Brownian motion (fBm). Leveraging the occupation time formula and Fernique's Theorem, we derive exact parametric restrictions on bandwidth and sampling rates ensuring convergence. We validate these theoretical findings via extensive simulations evaluating the mean squared error under varying noise persistence. Crucially, we explicitly benchmark the performance of the estimator under random sampling times against classical deterministic grids, while assessing the relative efficiency of quartic versus Gaussian kernels. Finally, we introduce an interactive Shiny application to dynamically visualize the admissible parameter space, and we empirically validate our theoretical framework by applying the estimator to the Vostok ice core paleoclimatological dataset. This real-world application demonstrates the practical capacity of the methodology to extract macroscopic trends while robustly avoiding severe overfitting under random sampling times.\
Ulrich Hounyo, Jiahao Lin, Xiaojun Song, Hansheng Wang
Fri 13 Nov
econ.EM
pdf html
- reading nominal, gmm, uniformly, straightforward
- interest Semiparametric inference and debiased machine learning 0.84
- your work near “Valid Inference with Imperfect Synthetic Data” 0.79
Abstract
We establish uniformly asymptotically exact inference for the population mean of two-way clustered data, allowing the data-generating process to vary with sample size. This finding does not contradict the well-known impossibility result of Menzel (2021), which places a fundamental limit on uniform estimation of the asymptotic distribution of the normalized full-array mean. We estimate the same population mean by averaging the sample means from random matchings. A closed-form standard error makes the resulting confidence interval straightforward to implement. The variance estimator is positive semidefinite and the studentized matching estimator is uniformly asymptotically standard normal. The cost of uniform validity is a slower convergence rate under degeneracy. We extend the result to smooth functions of GMM parameters, allowing the different moment components to exhibit different dependence regimes. In simulations, the proposed confidence intervals achieve coverage close to the nominal level across the dependence regimes considered.
Tianyue Zhou, Susan Gruber, Hana Lee, Wonyul Lee, Lei Nie, Mark van der Laan
Fri 13 Nov
stat.ME
pdf html
- reading adjustment, functional, randomized, functional form
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.85
- favorites near “Bayesian Semiparametric Model for Sequential Treatment Decisions with Informative Timing” 0.83
Abstract
Covariate adjustment can potentially improve efficiency in randomized clinical trials, but in rare diseases, limited prior knowledge makes it difficult to specify an adjustment at the design stage. Targeted maximum likelihood estimation (TMLE) with the Super Learner (TMLE+SL) can outperform fixed adjustment methods, such as the G-computation estimator, by leveraging machine learning to learn both the covariates and the functional form of the adjustment from the trial data. Estimating the known treatment mechanism can improve efficiency further. In simulations reflecting real rare-disease trials, TMLE+SL yielded valid inference and achieved greater efficiency and power than conventional analyses. Finally, we also provide practical guidance on specifying TMLE+SL in rare-disease trials.
Mai Mai, Matias Quiroz, Minh-Ngoc Tran
Tue 17 Nov
stat.MEstat.CO
pdf html
- favorites near “Variational Bayes under Model Misspecification” 0.89
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.80
- interest Bayesian computation and structural estimation 0.83
Abstract
Bayesian variable selection under spike-and-slab priors provides interpretable posterior inclusion probabilities for individual predictors, but exploring large discrete model spaces using conventional Markov chain Monte Carlo methods can be computationally challenging. We develop a collapsed importance-weighted variational framework for Gaussian linear regression. The framework analytically integrates out the remaining model parameters and directly approximates the posterior over binary inclusion indicators using a factorised Bernoulli distribution. To improve posterior inclusion probability estimation while retaining the computational simplicity of the factorised variational family, we replace the conventional evidence lower bound with a tighter variational R\'enyi importance-weighted lower bound. We establish a finite-sample upper bound on posterior inclusion probability approximation errors and show that, under suitable conditions, the optimised variational inclusion probabilities converge to their exact posterior counterparts as the number of importance samples increases and the R\'enyi parameter decreases appropriately. We further establish that analytical marginalisation tightens the variational objective and the corresponding inclusion probability error bound, while reducing importance-weight variance and improving the asymptotic efficiency of importance-sampling post-processing. To address the challenges of estimating gradients with respect to the variational parameters when the inclusion indicators are discrete, we leverage recent advances in variance-reduced score-function estimators, together with an annealing strategy to stabilise optimisation. Simulation experiments and empirical applications demonstrate improvements in posterior inclusion probability estimation relative to standard variational inference and two existing variational approaches.
Ariel Linden
Tue 17 Nov
stat.AP
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.84
- reading adjustment, empirical, researchers, ad
- interest Causal inference and identification in econometrics 0.81
Abstract
Health researchers commonly use weighting methods to improve covariate balance when estimating program and intervention effects from observational data. Standardized mean differences (SMDs) and variance ratios assess particular features of balance but may overlook residual differences elsewhere in the covariate distribution. This paper provides practical guidance for using Kolmogorov-Smirnov (KS), Anderson-Darling (AD), and Cramer-von Mises (CVM) statistics as complementary distributional balance diagnostics. In an empirical evaluation of a disease management program, weighting based on a main-effects propensity-score model produced SMD and KS statistics below commonly used or proposed balance thresholds of 0.10 for two continuous covariates, which would conventionally be interpreted as indicating acceptable balance. Nevertheless, permutation tests identified residual distributional differences. Adding quadratic terms for these covariates approximately halved their KS statistics, improved variance balance, and eliminated evidence of residual distributional differences, while leaving the estimated program effect substantively unchanged. Supporting simulations showed that KS, AD, and CVM differed in sensitivity to different forms of distributional discrepancy and that the calibration of resampling-based p-values depended on how the reference distribution was constructed. These findings support a practical approach that combines conventional balance measures, graphical assessment, complementary distributional diagnostics, weighting-model refinement, and outcome adjustment when important residual imbalance remains.
Qian Qin
Tue 17 Nov
math.STstat.TH
pdf html
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.80
- interest Bayesian computation and structural estimation 0.84
- reading mcmc, heavy-tailed targets, relate, inequalities
Abstract
We establish quantitative mixing-time bounds for Markov chain Monte Carlo algorithms targeting multimodal and heavy-tailed mixtures. The dynamics of a Markov chain are decomposed into local and global levels, and possibly weak Poincar\'{e} inequalities at both levels are combined to form convergence bounds. We apply this framework to location and scale mixtures of strongly log-concave distributions, obtaining explicit convergence bounds for random-walk Metropolis, proximal, and random-scan Gibbs samplers. For location mixtures connected along a finite path, the bounds depend polynomially on the number of components and exponentially on neighboring separation, remaining informative as the overall diameter grows. For scale mixtures, we relate the global dynamics of the chain to the isoperimetry of the log-scale mixing distribution, and illustrate the effect of heavy-tailed targets by examining inverse-gamma mixing distributions.
Mame Diarra Fall, Éric Barat, Nicolas Dobigeon
Mon 16 Nov
stat.MLcs.LGmath.OC
pdf html
- favorites near “Variational Bayes under Model Misspecification” 0.88
- interest Semiparametric inference and debiased machine learning 0.84
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.79
Abstract
Regular exponential-family distributions are ubiquitous in statistics due to their appealing and convenient properties. Notably, they can be expressed in terms of Bregman divergences, revealing fundamental connections between statistics, convex analysis, and information geometry. However, this geometric structure has not been systematically exploited to characterize these distributions with respect to their expectation parametrization. Building on a projective characterization, we establish a general construction of equivalent dual Bregman representations, together with normalization and uniqueness results. These representations express log-densities as ordinary Bregman divergences in dual coordinate systems, providing geometric characterizations of the parameter space. So-called $\beta$- and $\alpha$-scaled dual representations emerge as particular instances of this framework. We further derive dual Bregman representations of conjugate priors and establish that they preserve conjugacy. Applications to classical distributions illustrate the geometric structures underlying seemingly different statistical models. Moreover, we show that these representations provide natural reference geometries for establishing relative smoothness of the primal Bregman representation of the distribution, with explicit bounds that are sharp in the scalar and separable settings. An illustrative application to the Poisson distribution sheds new light on several works in the literature, providing complementary geometric justifications for reference geometries while deriving existing algorithms.
Mojtaba Soltanalian
Mon 16 Nov
cs.ITcs.LGmath.ITstat.ML
pdf html
- reading theory, infinite, state, govern
- interest Computational methods for heterogeneous agent and dynamic economic models 0.78
- your work near “Automated Solution of Heterogeneous Agent Models” 0.81
Abstract
An information system often has to learn before it knows what it will be asked. A finite representation can succeed without recording an infinite environment in full when those asks share collective structure, rather than demand unrelated facts scattered arbitrarily across it. We formalize this phenomenon through future-task information, the minimum retained information needed to preserve decision value relative to the available evidence. A joint decision-regret state records posterior comparisons among actions for every permitted task. Under compactness assumptions, every retained representation admits a calibrated replacement based on this state, preserving all task regrets without increasing information; the resulting information-regret curve equals an operational coding rate. Two exact boundaries govern finite preservation: weak-star closure for a fixed action count, and the singular mass of the source-evidence law when all finite action menus are admitted. Collective structure determines information requirements beyond individual-question difficulty. Families with identical source laws, pairwise task metrics, and individual-task rates can have different high-resolution exponents. Covariance-identical coordinate and parity families have different exact rate-distortion curves, with optimal parity retention characterized by a sharp Fourier-entropy inequality. For balanced nested task families, refinement sizes and weighted reach determine the rate within an additive constant, attained by prefix coding under explicit decay conditions. Nonlinear learning and computational separations distinguish preservation from efficient discovery and use. The proposed mathematical theory identifies which shared decision requirements allow finite information to remain useful for infinitely many possible purposes.
Jennifer Zhang, Amanda Coston
Thu 19 Nov
cs.CYcs.LGstat.ME
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.80
- reading science, llms, affect, questions
- interest Language models as statistical objects 0.80
Abstract
Generative AI is increasingly used to assist with data science tasks, yet fundamental questions remain about its ability to correctly formulate analytical problems. We investigate how well large language models (LLMs) can formulate prediction problems when developing models for high-stakes decision making. We focus on one particular challenge to validity: decision dependency, which arises when historical decisions affect observed outcomes. We evaluate this aspect of problem formulation in LLM-generated code and explanations across child-welfare, healthcare, and employment settings, using 11 models from OpenAI and Anthropic and seven prompt conditions. We assess whether the LLM-generated code implements a procedure appropriate for estimating the intended outcome under the study's assumptions and whether the explanation acknowledges decision dependency. We find that LLMs do not reliably address decision dependency without guidance in the prompt. When prompts specify the target, the assumption needed to estimate it, and the estimation procedure, target-aligned implementation reaches 99-100%. Among frontier models, a simple prompt description of how decisions affect outcomes appears sufficient to elicit target-aligned implementation. Some responses also acknowledge decision dependency without addressing it in their code. Our findings suggest that analysts should not rely on LLMs to correctly formulate prediction problems without explicit methodological guidance and careful review. We propose validity-centered guidelines for LLM-assisted predictive modeling, covering prompt formulation, validation of the target and estimation procedure, and identification of key assumptions. Establishing validity is a necessary early step in evaluating whether to deploy a predictive model in a high-stakes setting, but deployment requires weighing broader ethical, institutional, and practical considerations.
Felipe Angelim, Alessandro Leite
Thu 19 Nov
stat.MEcs.LG
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.87
- reading doubly robust, causal, treatment, doubly
- favorites near “Distributional Random Forests: Heterogeneity Adjustment and Multivariate Distributional Regression” 0.84
Abstract
Balancing weights are used in observational causal inference to construct a population in which the treatment is independent of observed pretreatment covariates. The ideal weight is the ratio between the product of the covariate and treatment marginals and their joint distribution, for binary and continuous treatments alike. We propose Balancing Partition Tree (BPT) and Balancing Partition Forest (BPF), which estimate this ratio on a learned rectangular partition of covariate--treatment space. Each split maximizes the gain in a Bregman divergence computed from joint and marginal counts. Two choices of divergence yield criteria that maximize the mutual information captured by the partition or reduce the squared error of the weights. We show that the population residual of either criterion bounds the bias of inverse-probability-weighted and doubly robust response estimators that use the population partition weights. Across 17 synthetic and semi-synthetic settings and three response estimators, BPF obtains the lowest mean RMSE rank among seven weight estimators, and a single tree fits in 0.14 seconds on 16,512 observations.
Houtian Yi, Zhaolun Wang, Xin Zhao
Tue 17 Nov
stat.ME
pdf html
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.83
- favorites near “The Minimax Regret of Sequential Probability Assignment, Contextual Shtarkov Sums, and Contextual Normalized Maximum Likelihood” 0.84
- interest Adaptive data collection and anytime-valid inference 0.82
Abstract
Streaming datasets require statistical models to be updated as new observations arrive, often without repeated access to the full history. In generalized nonparametric regression, a fixed spline space may limit approximation accuracy when the initial representation is insufficient. However, when the basis is expanded, the historical summaries represented in the previous spline space must be re-expressed in the enlarged spline space, which requires a suitable mechanism. We propose a renewable penalized-likelihood algorithm for canonical exponential-family nonparametric regression that combines nested cubic-spline refinement with a Gram-metric pullback of the retained historical likelihood quadratic under the streaming setting. The algorithm incorporates incoming-batch likelihood contributions and updates historical score and curvature summaries at the actual fitted center. The Gram-metric pullback preserves the retained historical quadratic over the embedded previous space. Under explicit regularity and algorithmic conditions, we establish L2-consistency at rate O_p(N^{-2/5}) for a fixed target in H^2[0,1]. We further establish that the energy-norm discrepancy between the renewable estimator and a matched offline estimator is of smaller order, where the offline method is fitted based on the full data and using the same realized terminal spline space and penalty. Simulations demonstrate the benefits of dynamic bases when the initial spline representation is restrictive, with additional gains from historical summary pullback. Additional experiments assess computational cost and storage of the proposed method relative to the offline method. Applications to real datasets also illustrate the proposed method.
Tobias Höppe, Stefan Bauer, Qiang Liu, Andrea Dittadi, Kirill Neklyudov
Tue 17 Nov
stat.MLcs.LG
pdf html
- favorites near “Sequential Monte Carlo Steering of Large Language Models using Probabilistic Programs” 0.80
- your work near “Timing as an Action: Learning When to Observe and Act” 0.77
- interest Language models as statistical objects 0.81
Abstract
Few-step generators via flow models require complicated technical solutions such as a pretrained flow model or a consistency loss. While rectified flows have gained popularity for their success in few-step sampling, they pay a cost of needing a pre-trained model and sampling a large number of couplings per dataset. Yet for a finite dataset and a Gaussian probability path, the probability-flow vector field of the empirical distribution is available in closed form, so one could use this field as the teacher and skip pretraining altogether. We demonstrate that, surprisingly, this approach is impossible for problems of practical relevance. In particular, the analytic solution of the diffusion process in high dimensions becomes almost a linear transformation. The integration returns the noise vector drawn at initialisation, plus a drift that is empirically a faint copy of the data point, almost independent of the noise. The coupling therefore differs from an independent pairing only through this drift. As a result, training a model on these couplings is almost equivalent to training the model on the independent draws of the noise and the data, which does not provide a useful teacher. We highlight that this is the direct consequence of the high-dimensionality of the data, and that for low enough dimensions or large enough datasets, analytic couplings become more efficient teachers. We quantify this effect by the fraction of endpoint energy carried by the drift, which scales as $\frac{\log N}{d}$ for a dataset of size $N$ in a $d$-dimensional space. Finally, we provide an empirical study confirming all the theoretical results presented in the paper.
Vicente J. Bolós, Mireia Mollar-Gumbau, Aleix Alcacer, Irene Epifanio
Tue 17 Nov
stat.MEstat.AP
pdf html
- interest Semiparametric inference and debiased machine learning 0.85
- reading business, special, functional, implement
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.78
Abstract
In this work, we introduce subarchetypal analysis, a new semi-supervised framework for Archetype Analysis (AA) and Archetypoid Analysis (ADA) based on subsets of observations from which archetypal profiles are generated. To the best of our knowledge, this is the first subset-constrained framework for archetypal analysis. Classical AA and ADA are recovered as special cases, and the location result for archetypes is extended to the proposed framework. The generating subsets may be defined a priori by the user to incorporate external knowledge, or determined from intrinsic properties of the data or the underlying problem structure. As an illustration of the latter, we consider efficient subarchetypes within Data Envelopment Analysis, where the subsets are determined by the geometry of the efficient frontier. We develop and implement algorithms for subarchetypal analysis in both multivariate and functional settings, available in an R package. The proposed methods are compared with constrained clustering, group-wise archetypal analysis, and classical unconstrained AA and ADA, and their practical usefulness is demonstrated through applications to business, sports analytics, and human development data.
Renato Paes Leme, Rad Niazadeh
Fri 13 Nov
cs.DScs.GTcs.LGmath.PRstat.CO
pdf html
- reading doubly, gives, far, cube
- interest Adaptive data collection and anytime-valid inference 0.83
- favorites near “Eluder Dimension and the Sample Complexity of Optimistic Exploration” 0.83
Abstract
We provide an efficient implementation of combinatorial Bernoulli factories. Such a factory turns coins of unknown biases $x$ into a random vertex of a polytope $P$ whose mean is exactly $x$. For every affine slice of the cube (the class of polytopes that admit a factory), we give a single algorithm whose expected number of coin flips and running time at every interior $x$ are polynomial in the dimension and in $1/{\delta_*}$, where ${\delta_*}=\min_i\{x_i,1-x_i\}$ is never supplied to the algorithm. For perfect matchings this gives $O(n^{8}{\delta_*}^{-2}\log(n/{\delta_*}))$ flips, while the factory of Niazadeh, Paes Leme, and Schneider [STOC 2021] spends at least $n^{2n-1}$ flips in expectation at the uniform doubly stochastic matrix. For other polytopes, our method rounds exactly given a point deep inside $P$ and the promise that $x$ is far from the boundary of $P$.
Sihwa Park, Jingsen Zhu, Vinamr Jain, Sheng-Yen Chou, Alexander Terenin
Fri 13 Nov
cs.LGcs.AIstat.ML
pdf html
- reading functions, process priors, gaussian process, randomized
- favorites near “Fit without fear: remarkable mathematical phenomena of deep learning through the prism of interpolation” 0.82
- interest Kernels, operators and functional data 0.80
Abstract
Thompson sampling is natural randomized strategy for optimizing unknown functions in a black-box manner. In continuous global optimization, Thompson sampling has largely been restricted to Gaussian process priors, as these are essentially the only class for which computing the posterior distribution has been tractable in practice. In this work, we introduce \emph{prior-fitted diffusion transformers}, which are diffusion models that perform Thompson sampling in-context at inference time, after being pre-trained using the prior-fitted network paradigm. To apply these models, we develop a set of practical non-Gaussian priors for general black-box functions---including those that have characteristics that cannot be achieved under Gaussianity, such as priors over unimodal functions. We also quantify the tradeoff between model size and the number of pre-training iterations, and show that these models follow scaling laws akin to those observed in transformers defined over other modalities. Across a set of global optimization benchmarks, we show that prior-fitted diffusion transformers with non-Gaussian priors can achieve improved performance compared to traditional pipelines built on Gaussian processes.
Dong-Sig Han, Tolga Birdal
Mon 16 Nov
cs.LGstat.ML
pdf html
- reading theory, hamiltonian, dynamical, complex
- favorites near “Hypoelliptic diffusion maps I: tangent bundles” 0.86
- interest Kernels, operators and functional data 0.82
Abstract
Diffusion bridges have emerged as versatile frameworks for facilitating transformations between arbitrary data distributions. However, these approaches are hindered by an immense optimization space and its underlying high-order structures. This paper explores a variant of diffusion bridges for solving real-world problems in complex vector spaces governed by scaling and rotation laws, such as Hamiltonian structures. To address this, we introduce conformal optimal transport (COT), a complex-valued transportation framework designed for physical systems with geometric and dissipative dynamics under a geometric lifting. Building on this, we propose a new class of dynamical diffusion bridges, called complex Schr\"odinger bridges (CSBs), whose derived algorithms can handle diffusion-based tasks using a conformal geometric prior. For scalability, we propose image-to-image CSB (I$^2$CSB), a tractable diffusion-bridge technique with an analytic spiral posterior. Finally, we implement a physics-informed CSB algorithm within simulations of quantum electrodynamics, mimicking the logical behavior of quantum logic gates. Our extensive results validate the underlying theory while highlighting advantages of complex-valued diffusion.
Yikun Li, Matey Neykov
Mon 16 Nov
math.STstat.TH
pdf html
- interest Adaptive data collection and anytime-valid inference 0.84
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.80
- favorites near “Orthogonal Statistical Learning” 0.87
Abstract
We study nonparametric regression over star-shaped function classes under strong adversarial contamination. The clean (original) observations $(\tilde{X} _{i},\tilde{Y} _{i})_{i=1} ^{n}$ are i.i.d. and satisfy $\tilde{Y} _{i}=f ^{*}(\tilde{X} _{i})+\xi _{i}$, where $f ^{*}\in \mathcal{F}$, $\tilde{X}$ is distributed as $\mathbb{P} _{X}$, and the noise is conditionally centered. An adversary may inspect the clean sample and replace at most $\lfloor \epsilon n\rfloor$ observation pairs. We assume that $\mathcal{F}\subset L _{2}(\mathbb{P} _{X})$ is star-shaped with finite diameter $d$. Under a Bernstein-type moment condition on function increments and a bounded conditional second moment for the noise, we establish the matching lower and upper bounds of order $\left((\eta ^{*}) ^{2} \vee \epsilon \right)\wedge d ^{2}$, uniformly over admissible star-shaped function classes, therefore characterizing the minimax rate of estimation error, where \begin{equation*} \eta ^{*}:=\sup\left\{\eta >0: n\eta^2\leq\log \mathcal M _{\mathcal{F}}^{\text{loc}}(\eta,c)\right\}. \end{equation*} Here $\log \mathcal{M} _{\mathcal{F}} ^{\text{loc}}(\eta,c)$ denotes the local metric entropy of $\mathcal{F}$ in $L _2(\mathbb{P} _{X})$, and $c$ is a sufficiently large constant. Under a uniform relative sub-Gaussian condition on function increments and noise, the lower bound is adjusted to $\left((\eta ^{*})^{2}\vee \epsilon ^{2}\right)\wedge d ^{2}$ while the upper bound improves to $\left((\eta ^{*})^{2}\vee \epsilon ^{2}\log ^{2}\left(\frac{e}{\epsilon }\right)\right)\wedge d ^{2}$, yielding minimax optimality up to logarithmic factors.
Noonhil Kizhakkemadam Unnikrishnan
Tue 17 Nov
stat.MEmath.STstat.APstat.TH
pdf html
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.80
- favorites near “Optimal uniform convergence rates and asymptotic normality for series estimators under weak dependence and weak conditions” 0.87
- interest Bayesian computation and structural estimation 0.83
Abstract
We develop Bayesian inference for multivariate $\alpha$-stable distributions with von Mises--Fisher spectral measure, motivated by the hitting distribution of Brownian motion on the unit sphere. Inference proceeds via a LePage-type series representation with latent radial, spectral, and Poisson components; truncation bias is absorbed through a Gaussian error term, yielding a likelihood that is smooth in $\alpha\in(0,2)$ with no special treatment at the Cauchy boundary $\alpha=1$. The stability index $\alpha$, spectral concentration $\kappa$, mean direction $\eta$, and spectral mass $\Gamma_\infty$ are estimated jointly by Markov chain Monte Carlo. At the Gaussian boundary $\alpha\to 2$, the limiting covariance matrix equals twice the second moment matrix of the spectral measure, with explicit parametric form determined by $\eta$ and $\kappa$; in particular, the anisotropy of the limiting Gaussian distribution is controlled solely by the spectral concentration $\kappa$. Simulation across $\alpha\in[0.25, 1.99]$ confirms reliable recovery of all parameters.
Hengtao Zhang, Yuanke Qu, Mi-Ok Kim, Huaqing Jin
Tue 17 Nov
stat.ME
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.85
- reading borrowing, randomized, historical, control
- interest Causal inference and identification in econometrics 0.78
Abstract
Incorporating historical controls into randomized clinical trials can improve statistical efficiency and alleviate enrollment challenges by reducing the need for concurrent controls, particularly when their recruitment is ethically or logistically challenging. These gains depend on comparability between the historical and concurrent controls. Many dynamic borrowing methods assess compatibility for the overall study populations and apply the same borrowing decision to every participant, even when comparability varies with participant covariates. We propose RADISH, Robust Adaptive Discrepancy-Informed Shrinkage for Historical Borrowing, a covariate-adaptive design that assesses compatibility at the individual participant level and incorporates robust borrowing into sequential randomization and final inference. Under the stated regularity conditions, the RADISH estimator is asymptotically normal and supports asymptotically valid Wald inference. Simulation studies, including semi-synthetic simulation studies based on the two real-world randomized trials, show that RADISH improves efficiency and assigns fewer participants to the control group when the historical and concurrent controls are compatible. As incompatibility increases, RADISH progressively reduces borrowing and approaches a concurrent-only design and analysis.
Christian E. Galarza, Víctor H. Lachos, Dashun Liu
Mon 16 Nov
stat.ME
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.80
- interest Semiparametric inference and debiased machine learning 0.85
- reading lasso, written, lognormal, functions
Abstract
Modal regression models the conditional mode rather than the conditional mean, which is attractive for positive and right-skewed responses whose mean is distorted by the upper tail. We propose a regularized version of parametric modal regression for positive responses with high-dimensional covariates, in which the density parameters of the Gamma, Weibull and Lognormal distributions are written as functions of the conditional mode and the log-likelihood is penalized by the Lasso, adaptive Lasso or elastic-net penalties. The estimates are computed by a proximal Newton algorithm, equivalently an iteratively reweighted penalized least-squares scheme, whose inner problem is solved exactly by the coordinate-descent routines of glmnet, and the regularization parameter is selected by the extended Bayesian information criterion. We show that the Lognormal estimator coincides with the Gaussian Lasso applied to the log-transformed response, that under constant dispersion the conditional mean and the conditional mode have the same slopes, and that the estimation error attains the rate $O(\sqrt{s}\lambda_n)$ with no smoothing bias. Simulation studies with up to $p = 1000$ covariates show that the method improves on the Lasso applied to the log-transformed response for Gamma and Weibull responses, and two applications illustrate its use. The method is implemented in the R package ModalNet.
Hristo Papazov, Francesco D'Angelo, Nicolas Flammarion
Tue 17 Nov
cs.LGcs.AIcs.FLstat.ML
pdf html
- favorites near “Spectral State Space Models” 0.81
- your work near “Timing as an Action: Learning When to Observe and Act” 0.79
- interest Language models as statistical objects 0.80
Abstract
We introduce a framework for differentiable exact learning that couples an unbounded external environment to a state-tracking neural controller, which we train on policy-trajectory observations (PTOs): expert behavior traces that expose observations and actions while hiding internal states. In addition to the learning framework, we introduce Differentiable Finite-State Transducers (DFSTs) which use multilinear updates, support log-parallel training, and emulate arbitrary finite-state controllers, yielding computational universality through environmental interaction. From tiny training datasets, compact DFSTs achieve extreme length generalization on arithmetic and universal-simulation tasks. We replicate the success of DFSTs on RNNs and GRUs, thereby supporting a broader learning recipe: State-Tracking Controller + Unbounded External Environment + PTOs = Extreme Length Generalization. Beyond length generalization, our metric-extraction protocol recovers interpretable discrete controllers from trained DFSTs, which we prove equivalent to the target arithmetic and universal-simulation algorithms. Furthermore, for our Softmax DFST models, simply taking $\mathrm{argmax}$ over the transition probabilities yields discrete controllers which provably exactly perform universal Turing-machine simulation and binary arithmetic. Proving exact learning for the neural controllers requires assumptions, which we empirically support. We make our code, training recipes, model weights, and proof checkers publicly available at https://github.com/fanagor/DEL.
Suramya Surendran, Rajesh G
Wed 18 Nov
math.STstat.TH
pdf html
- interest Semiparametric inference and debiased machine learning 0.85
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.80
- favorites near “Eigendecompositions of Transfer Operators in Reproducing Kernel Hilbert Spaces” 0.83
Abstract
In this paper, a bivariate extension of the differential Renyi Tsallis divergence is introduced for comparing two absolutely continuous bivariate probability distributions. Fundamental properties of the proposed divergence are studied, including non negativity, pseudo additivity under component wise independence, and invariance under monotone and nonsingular linear transformations. Relationships with Renyi, Tsallis, alpha, Bhattacharyya, and Hellinger divergence measures are also established. A sensitivity analysis is performed to compare the Renyi, Tsallis and Renyi Tsallis divergences under the Marshall Olkin and Gumbel type bivariate exponential models, taking the magnitude of Kendall's rank correlation coefficient as a common scale for dependence. The proposed measure is further developed for bivariate order statistics, with particular emphasis on the joint distribution of the minimum and maximum order statistics for several lifetime distributions, including exponential, Weibull, Pareto, and Rayleigh models. When the underlying distributions are unknown, a nonparametric estimator is constructed using bivariate Gaussian kernel density estimators with plug in bandwidth selection. The proposed estimator is applied to a real life dataset and bootstrap resampling is used to assess estimation uncertainty. The results show the applicability of the proposed divergence for comparing bivariate distributions and indicate that its finite sample performance depends on the order parameter and the distributions being compared.
Rathindra Nath Karmakar, Samriddha Lahiry, Koichi Yamagata
Tue 17 Nov
math.STmath.MPstat.TH
pdf html
- favorites near “Introduction to regularity structures” 0.87
- interest Semiparametric inference and debiased machine learning 0.85
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.78
Abstract
In classical statistics, local asymptotic normality provides a general framework for characterizing optimal testing power and constructing asymptotically efficient procedures. Extending this framework to quantum statistical models introduces additional challenges arising from the noncommutativity of observables and the need to optimize over quantum measurements. In this paper, we develop a theory of asymptotic efficiency for quantum hypothesis testing using quantum local asymptotic normality and the asymptotic representation theorem. We establish sharp nuisance-maximin power bounds for one-sided tests and two-sided tests within the unbiased class by reducing the corresponding classical--quantum Gaussian shift experiments to classical testing problems. This reduction is justified through a quantum Hunt--Stein argument that accommodates both commuting and noncommuting Gaussian coordinates. Under suitable asymptotic invariance conditions, the resulting bounds transfer to the original quantum experiments and are attained by scalar symmetric logarithmic derivative tests. We further develop a two-stage adaptive construction that achieves the same local asymptotic power without knowledge of the reference parameter. As an application, we study testing of smooth spectral functionals of quantum states. We compare scalar symmetric logarithmic derivative tests with the weak Schur sampling plug-in test and with optimally weighted linear statistics of its fluctuations. This comparison distinguishes loss from the plug-in weighting from a gap between the best such spectral statistic and the scalar score procedure. We characterize equality of the corresponding variances and illustrate the two gaps through explicit examples.
Zhou Wang, Xingye Qiao
Tue 17 Nov
stat.MLcs.LG
pdf html
- favorites near “Preference-based Online Learning with Dueling Bandits: A Survey” 0.87
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.80
- interest Online learning and statistical learning theory 0.84
Abstract
In high-stakes environments where uncertainties abound, set-valued prediction offers a cautious and robust mechanism by presenting multiple potential labels as the prediction for each test instance to mitigate the potential risk associated with prediction errors. Yet, integrating this paradigm with out-of-distribution (OOD) detection remains scarcely explored in such settings as online learning with bandit feedback. The bandit feedback mechanism informs the learner about the correctness of the pulled arm/action instead of the explicit ground truth label, leaving the true class label unknown when an incorrect action is taken. To address this challenge, we introduce BanditGPS which conducts set-valued prediction with OOD detection in the bandit feedback setting, using an estimation to the ground truth of class labels. BanditGPS achieves three objectives: render small/informative prediction sets, enhance the OOD detection performance, and control the recall for all normal classes to meet prescribed requirements. Our approach is characterized by the loss function, which trades off between high OOD detection and small prediction sets. Theoretically, we prove that the convergence rate of the regret is $\tilde{\mathcal{O}}(T^{-1/2})$. The empirical results further show that BanditGPS effectively controls the recalls with promising performances on OOD detection and informative prediction.
Davide Wiest
Fri 13 Nov
cs.LGstat.ML
pdf html
- favorites near “Spectral State Space Models” 0.83
- interest Kernels, operators and functional data 0.82
- reading state, differentiation, held, modern
Abstract
Self-supervised objectives such as VICReg learn representations by making related inputs similar, while variance and covariance penalties prevent collapse. These penalties are statistics over a batch of samples, so layer-local variants, too, are typically local in space but not in time. We introduce Temporal Regularized Learning (TeReL), which optimizes a VICReg-style objective one sample at a time with purely local updates: each update uses only the current activity of its own and the preceding layer and a fixed-size running state. Temporally adjacent inputs serve as positive pairs, which makes the objective a soft form of Slow Feature Analysis (SFA). Running estimates of each neuron's mean and variance, together with one lateral matrix per layer, replace the batch statistics; held fixed during differentiation, they turn the gradient into a Hebbian feedforward update and an anti-Hebbian lateral update. We show that a two-layer network trained with batch size one on an MNIST stream of class-homogeneous chunks reaches 96.48% linear-probe accuracy, above random features (95.09%) and classical SFA, and that the temporal order and each loss term contribute to this result. With both locality constraints relaxed, the same objective matches supervised backpropagation. TeReL thus shows that the collapse-preventing statistics of modern self-supervised learning can be held as running state at each neuron, which brings these objectives closer to what brains and neuromorphic hardware can implement.
Frank M.T.A. Busing
Mon 16 Nov
stat.MLstat.CO
pdf html
- reading absolute, multidimensional scaling, procedures, locations
- favorites near “Monte Carlo Geometry Processing” 0.79
- interest Kernels, operators and functional data 0.81
Abstract
Node localization aims to estimate the positions of nodes based on collected data from other nodes. Depending on the data characteristics, different procedures are utilized to estimate the locations. In this paper, we introduce a comprehensive multidimensional scaling algorithm based on iterative majorization that comprises many of the different data characteristics to estimate node locations. The implicitly normalized loss function provides absolute distances and, given anchors, absolute node locations. The algorithm uses a new convergent coordinate descent procedure for the minimization of the majorization function, allowing for a natural treatment of anchors. Optimal transformations of the data seamlessly allows other data than Euclidean distances. Data from multiple sources, resulting in either asymmetric or three-way data, fit the same extensive framework.
Felipe de Jesús Félix-Arredondo, Nezih Nieto-Gutiérrez, Carlos Astengo-Noguez, Alejandro Ucan-Puc
Fri 13 Nov
math.OCstat.ML
pdf html
- favorites near “Recent Developments in Machine Learning Methods for Stochastic Control and Games” 0.82
- interest Computational methods for heterogeneous agent and dynamic economic models 0.78
- reading gains, tuned, controller, generic
Abstract
A discrete PID controller is exactly a recurrent neural network, a classical state-space fact: a fixed linear memory of the running integral and the last error, read out by three trainable weights that map one-to-one to the PID gains. Networks with bounded hidden states, such as LSTMs of any width, only approximate it. Trained only on a one-step physical cost (the expected cost of one transition from a fixed state distribution), with no demonstrations, teacher controller, or known gains, the network solves a strongly convex least-squares problem, so gradient descent converges from any initialization to the unique one-step optimal PID gains; the limit network is that PID, with the same action on every error sequence. Partitioned regression then shows what extra capacity buys. Orthogonal residual features keep these gains and improve on the PID if and only if they correlate with the part of the optimal action outside the PID span. On a nonlinear second-order example the gap is 5.88% of the optimal PID cost, and 4.00% against PID with a bias; no controller can improve on PID by more on this objective. A residual started at a tuned PID begins exactly at that PID; at the one-step optimal PID, a fixed-feature residual's first gradient step lowers the cost if and only if its features can improve on the PID. Experiments confirm the realization, recovery, and gap results to numerical precision. On held-out plants of three families, under a protocol frozen before the test data existed, a residual warm-started at a heavily tuned PID and trained with it lowers the long-horizon cost by 3.5%, 3.5%, and 1.9%, with family-wise 95% bounds excluding zero, also under distribution shift; a post hoc comparison with the PID tuned alone suggests that most of it comes from the recurrent part in two families. The primary endpoint, a gain of at least 5% from a generic PID start, is not met in any family.
Troy Larsen, Alen Alexanderian
Wed 18 Nov
stat.ME
pdf html
- interest Semiparametric inference and debiased machine learning 0.86
- favorites near “Operator-valued Kernels for Learning from Functional Response Data” 0.86
- reading high-dimensional models, kernel-based, nonlinear, formula
Abstract
We consider global sensitivity analysis of models with high-dimensional input parameters using HSIC-based indices. The numerical estimation of these kernel-based indices is known to be challenging for such models. Without proper kernel tuning, the standard estimator for the HSIC can become unstable and break down. Focusing on models with independent parameters, we introduce a dimension-aware bandwidth formula and provide theoretical analysis justifying the proposed approach. In particular, we show that unless kernel bandwidth parameters explicitly scale with the input dimension, the corresponding sensitivity indices return uninformative rankings of parameter influence. Further, we demonstrate the effectiveness of the proposed scaling strategy on linear and nonlinear test problems, including models with correlated inputs.
Xiaoyu Li, Jiangxuan Long
Tue 17 Nov
cs.LGcs.CLstat.ML
pdf html
- favorites near “The Minimax Regret of Sequential Probability Assignment, Contextual Shtarkov Sums, and Contextual Normalized Maximum Likelihood” 0.86
- interest Online learning and statistical learning theory 0.83
- reading retrieval, theory, data analysis, price
Abstract
In-context learning is usually analyzed as if the examples in the prompt were sampled before the learner is chosen. In deployed systems they are not: examples are retrieved, filtered, re-ranked, edited, and repeatedly tested after seeing the query and the data store. We formalize this gap by treating prompt construction as adaptive data analysis through a conditional stochastic-kernel calculus. Our main quantity is the context capacity, the mutual information or approximate max-information between the data store and the final prompt transcript. For any frozen in-context learner and any adaptive context pipeline with capacity $\kappa$, we prove a finite-sample context generalization bound of order $\sqrt{\kappa/k}$, where $k$ is the number of selected examples whose losses form the reported score. We prove a matching lower bound, whose constant is exact when $\log k\ll\kappa\ll k$, and derive two PAC-Bayesian retrieval rules: a first-order Gibbs rule that charges data-dependent candidate gates their prior mass, and a self-normalized second-order rule. On Banking77 with 64 retrieval recipes and two model families, the certificates hold in all 24,000 resampled selection events, and from 64 (Qwen2.5-7B) or 128 (GPT-4.1-mini) reused validation queries the Gibbs rule certifies a risk below the zero-shot error in over 99% of validation draws. Finally, we instantiate the theory for Bayesian linear in-context regression, including an order-statistics analysis of query-aligned top-$k$ Gaussian retrieval. The results separate context quality from context certification: longer or more relevant contexts improve risk only when the procedure used to choose them is stable enough for selected-context evidence to transfer.
Eman Kabbas, Janet van Niekerk, Håvard Rue
Fri 13 Nov
stat.ME
pdf html
- favorites near “Monte Carlo Geometry Processing” 0.82
- interest Semiparametric inference and debiased machine learning 0.84
- your work near “Automated Solution of Heterogeneous Agent Models” 0.80
Abstract
We introduce the Proper Random Walk (PRW) family of spline models with a particular focus on the Second-Order Proper Random Walk (PRW2), a proper stationary alternative to the standard random walk of order 2 (RW2) spline. The PRW2 retains the computational strengths and spline-like smoothing of the RW2 while providing uniform marginal variance and stable uncertainty behavior. The PRW2 model is governed by a range parameter that controls correlation decay: in the infinite-range limit, it recovers RW2. In addition to its discrete representation, we show that the PRW2 model admits a continuous-time formulation as the critically damped Mat\'ern model with smoothness $\nu=3/2$, which provides an interpretable parameterization and enables construction through a Galerkin finite-element approximation on general grids. To mitigate edge distortions on bounded domains, we propose a local boundary correction that enforces the stationary joint law while preserving sparsity and $O(n)$ factorization. Finally, we develop penalized complexity priors for the model hyperparameters, enabling principled shrinkage toward the RW2 limit. The model is applied to simulated and real data to illustrate its properties and advantages.
Shusen Ma, Yangchun Xie, Xudong Zuo, Haizhou Liang, Yun-Bo Zhao
Thu 19 Nov
cs.LGstat.ML
pdf html
- favorites near “Spectral State Space Models” 0.82
- reading history, forecasting, series forecasting, series
- interest Macroeconometrics and time series 0.79
Abstract
A fixed number of recent observations may span minutes for one variable and days for another. Earlier measurements can still reveal whether the latest value is unusual for that variable. We propose TRESS (Timestamped Recent Events and Statistical Summaries), a forecasting model that retains the latest observations at their original times and summarises the full history of each variable. For a given set of variables and event budget, this representation stays the same size as the record grows. Its query-conditioned event lifting module weights deviations from the latest value according to the age of each observation and the forecast horizon. A channel-specific decoder combines the resulting features with historical statistics and a summary shared across variables to predict a correction to the latest value. Retaining 8 to 16 observations per variable, TRESS records the lowest mean test MSE and MAE among the methods compared on all four benchmarks. Serving 32 successive query times from one encoded history reduces CPU inference time by a factor of 2.12--2.60 across the four datasets. The code is at https://github.com/SSMa913/NICLab-TRESS.
Kartik Pandit, Sourav Ganguly, Arnob Ghosh
Wed 18 Nov
cs.LGcs.AIstat.ML
pdf html
- reading controller, just, online, large
- favorites near “Recent Developments in Machine Learning Methods for Stochastic Control and Games” 0.78
- your work near “Timing as an Action: Learning When to Observe and Act” 0.78
Abstract
We study constrained contextual bandits, in which a learner chooses one of $K$ actions in each round to minimize loss while keeping the expected cost of its play below a budget. Existing reductions to online regression weight the cost by a Lagrange multiplier learned from past constraint residuals, so slack in some rounds can offset violations in others. We instead run SquareCB proposed in \citep{foster2020beyond} with a rectified weight on the cost: it is zero when the estimated cost is within the budget, and otherwise just large enough to bring it back to the budget. Under realizability and Slater's condition with margin $\delta$, the algorithm attains regret $O(\sqrt{KT\,\mathrm{Reg}_{\mathsf{Sq}}(T)}/\delta)$ and cancellation-free violation $\sum_t(V_c^{p_t}-b)^+=O(\sqrt{KT\,\mathrm{Reg}_{\mathsf{Sq}}(T)})$, where $\mathrm{Reg}_{\mathsf{Sq}}$ is the error of the regression oracle. The same bounds hold when the budget must be met in every context, potentially adversarially chosen contexts. In controller selection for a safe-control-gym quadrotor and a BOPTEST building, the violation is 6 to 9 times lower than that of primal-dual and virtual-queue baselines, with a lower or comparable regret.
Christian Bayer, Jakob Kellermann
Mon 16 Nov
math.STmath.PRstat.TH
pdf html
- reading functional, question, supplement, state
- favorites near “Operator-valued Kernels for Learning from Functional Response Data” 0.85
- interest Kernels, operators and functional data 0.83
Abstract
Path signatures are a standard tool for learning functionals on path space. The main underpinnings of their application are universal approximation results. For a wide range of function classes on path space, these results state that any function in such a class can be approximated arbitrarily well by a linear functional of the path signature. Hence, linear functionals of the signature are in principle expressive enough for learning functions on path space. Nonetheless, these results do not provide insights into the statistical question of how much data is needed to identify a suitable signature functional. In this work, we study this question in the framework of kernel ridge regression with the signature kernel. The target is the conditional expectation $\E[Y|X]$, where $X$ is the observed path and $Y$ is a real-valued response. Under different assumptions on the moments of the signature kernel and the response, we provide finite-sample and consistency guarantees for the method. We supplement our theoretical analysis with numerical experiments that illustrate our results.
Jiaxin Shi, Ziyuan Chen, Jian Huang, Hansheng Wang
Thu 19 Nov
stat.MLcs.LGmath.STstat.TH
pdf html
- favorites near “Neural Operator: Learning Maps Between Function Spaces With Applications to PDEs” 0.83
- interest Computational methods for heterogeneous agent and dynamic economic models 0.80
- your work near “Automated Solution of Heterogeneous Agent Models” 0.79
Abstract
Stability under input perturbations is necessary for deep neural network estimators in many applications involving noisy covariates or estimated features. Weight norm constraints are widely used to control sensitivity to input perturbations by bounding the Lipschitz constant of neural networks. However, existing approximation results under fixed bounds on standard norm products are insufficient to establish optimal statistical rates. In this paper, we establish a sharp approximation theory for deep ReLU networks under a split spectral norm product constraint. This theoretical result shows that sufficiently smooth H\"{o}lder functions can be uniformly approximated to arbitrary accuracy as the network depth increases, with the norm-product bound held fixed. We then propose a penalized least squares estimator for deep nonparametric regression. The resulting estimator has a uniformly bounded Lipschitz constant while still achieving the optimal nonparametric rate up to a logarithmic factor. As applications, we establish optimal prediction rates for two-stage learning and provide robustness guarantees under adversarial input perturbations. Numerical studies show that the penalized estimator has well-controlled sensitivity to input perturbations and prediction accuracy comparable to that of classical deep network methods.
Nhi Nguyen, Shauli Ravfogel, Rajesh Ranganath
Fri 13 Nov
cs.LGcs.AIcs.CLstat.ML
pdf html
- favorites near “Sequential Monte Carlo Steering of Large Language Models using Probabilistic Programs” 0.82
- reading llms, explains, versus, feedback
- interest Language models as statistical objects 0.81
Abstract
As large language models (LLMs) increasingly make consequential decisions and solve problems alongside humans, we must understand how models arrive at their conclusions. Understanding how an LLM behaves requires observing it across multiple inputs, yet free-text explanations such as chain-of-thought (CoT) are generated per instance and cannot be easily verified on new inputs. We propose {code explanations to address the mismatch in how LLM explanations are generated versus how they are evaluated: a program explains LLM behavior across a dataset of inputs, and its execution enables automatic verification. We define Consistency metric that measures the fraction of inputs on which the program reproduces the LLM's outputs, and use it to evaluate and select candidate programs. Across three open-source code-capable LLMs and eight tasks, we find that LLMs can produce highly consistent code explanations in several settings. Code explanations are also 6$\times$ to 100$\times$ shorter than CoT, and better align with model behavior in 9 out of 24 settings. However, consistency is significantly lower on tasks with more complex linguistic components, where generated programs reproduce model outputs no more frequently than an uninformative explanation baseline, and these programs often rely on longer hard-coded patterns. In experiments with Claude Opus 5.5, providing the model with consistency feedback and code execution improves explanation consistency across tasks, but consistency remains significantly lower on more complex tasks. Our findings establish executable code as a novel, verifiable, and compact approach to explaining LLM behavior and highlight verification feedback as a promising direction for improving explanation consistency. The code for this paper is at https://github.com/rajesh-lab/code-as-explanations.
Zengqing Wu, Chuan Xiao
Fri 13 Nov
stat.APcs.AIcs.CY
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.84
- interest Reinforcement learning, policy learning and off-policy evaluation 0.80
- favorites near “More power to you: Using machine learning to augment human coding for more efficient inference in text-based randomized trials” 0.83
Abstract
Survey simulations require realistic combinations of answers, but predictive accuracy alone does not establish that a model can generate them. We evaluate the option probabilities returned by Jev, a TypeSafe decision model, using linked responses to 20 binary items from five archived survey waves. We compare population-only prediction, prediction given recorded answers, and sequential completion that uses generated answers to predict later responses. Joint comparisons hold the starting population information fixed, and two three-question cases are completed in both question orders. We evaluate predictive scores, answer frequencies, and associations against the survey records. Recorded answers can improve predictive scores while associations remain distorted. Large conditional errors also occur. For example, among men who described themselves as having traditional values, the model predicted that 16% would see men and women as similar in workplace abilities, compared with 56% in the survey. Sequential completion has mixed effects on joint validity, and reversing question order produces different joint distributions in both tested cases. Order checks and classical marginal bounds clarify what can be assessed before linked validation answers are available. The findings support validating the joint events a simulation uses and separating empirical constraints from model assumptions. The evidence concerns selected binary tasks and one model version.
Jon Kleinberg, Fan Wei
Fri 13 Nov
cs.DScs.FLcs.LGstat.ML
pdf html
- favorites near “Sequential Monte Carlo Steering of Large Language Models using Probabilistic Programs” 0.85
- reading large, language, density, rectangles
- your work near “Valid Inference with Imperfect Synthetic Data” 0.76
Abstract
Large language models sharpen the validity--breadth trade-off: a generator should avoid hallucinations without collapsing onto a narrow set of valid outputs. We study generation from positive examples of an unknown target in the Kleinberg--Mullainathan model, with arbitrary, possibly adversarial order and no distributional or computational assumptions. Breadth is measured by lower Banach density, which rules out arbitrarily large sparse holes, unlike prefix-based lower asymptotic density. 1. Exact optima in one dimension: For each target $K$ in a countable family, we characterize its optimal lower Banach density $b(K)$ via equivalent combinatorial and topological formulations. One eventually valid generator attains all $b(K)$ simultaneously. Every $\alpha\in[0,1/2]$ occurs as $b(K)$ in some countable family; finite Cantor--Bendixson point rank guarantees $b(K)=1/2$, even without a uniform rank bound. 2. Unavoidable trade-offs in higher dimensions: For $d\ge2$, there need not be a single best generator: optimizing density for one target may worsen the guarantee for another. An explicit family quantifies this intrinsically multi-objective trade-off. 3. Complete characterization of density profiles: We characterize the full region of simultaneously achievable guarantees, with equivalent combinatorial and topological formulations, after excluding geometrically degenerate rectangles. Every prescribed product $\prod_i[0,\alpha_i]$, $\alpha_i\in[0,1/2]$ can occur as an exact achievable region for some language collection. 4. Near-half guarantees and the sharp endpoint: Under a near-logarithmic nondegeneracy condition, one generator achieves filtered density at least $1/2-\varepsilon$ at all finite-point-rank targets, for every $\varepsilon>0$, improving the previous polylogarithmic threshold. The exact $1/2$ endpoint can fail even for a rank-zero singleton family.
Ephraim Kofi Ayi Kwapong
Tue 17 Nov
cs.LGstat.ML
pdf html
- favorites near “Fit without fear: remarkable mathematical phenomena of deep learning through the prism of interpolation” 0.82
- reading good, summary, kind, explain
- interest Language models as statistical objects 0.81
Abstract
A network trained on task A and then task B ends at different weights than one trained on B and then A. To leading order this gap is set by the Lie bracket of the two tasks' gradient fields, and Dherin (2023) conjectured that a non-vanishing bracket may be linked to catastrophic forgetting. I test this conjecture empirically. Across Split MNIST, Split Fashion-MNIST and Split CIFAR-10, the forgetting of A is almost exactly the difference in A's loss between the two training orders (Spearman rho = 0.96-1.00). I then ask whether the bracket can predict that forgetting early in training on B and without any of A's data, using only a small summary stored when A finishes: its gradient and its top 20 Hessian eigenpairs. When the old task is well learned and forgetting is slow (task-incremental), this stored bracket ranks forgetting with rho = 0.70-0.81, against 0.33-0.58 for the penalty of Elastic Weight Consolidation, and matches or exceeds re-evaluating A on its own data. When forgetting is fast (domain-incremental), the stored summary goes stale and the EWC penalty is as good or better. The bracket fails outright when the old task is only half learned; I trace this to a residual gradient on A an order of magnitude larger, which makes first-order (slope) damage matter. The bracket is therefore a useful early-warning signal for the quiet kind of forgetting, under a clear condition.
Mrigank Shekhar Singh
Fri 13 Nov
stat.MLcs.AIcs.GTcs.LG
pdf html
- interest Adaptive data collection and anytime-valid inference 0.85
- your work near “Timing as an Action: Learning When to Observe and Act” 0.80
- favorites near “Recent Developments in Machine Learning Methods for Stochastic Control and Games” 0.75
Abstract
Recursive estimators on drifting streams must forget stale data fast enough to track the target, yet slowly enough to average out noise. In prior-anchored ridge regression, this trade-off admits a closed-form spectral decomposition in which each eigen-direction interpolates between the data solution and the prior with pull weight $s_i = \rho / (\mu_i + \rho)$. When the anchor is refreshed every $k$ steps, we show that the moving-frame offset converges to a finite-cadence equilibrium that lags the pure-leaky path by an explicit $\Delta_k(s)$ while reducing estimation variance by a factor $\Psi(s, r_k)$. Minimizing the cycle-averaged risk gives $\omega^* = (4\nu^2 \Sigma / c')^{1/3}$, $c' = \sigma^2 d_{\mathrm{eff}}(\rho)$, and the plug-in strength $\rho^* = \sigma^2 / (n\delta^2)$. A random-matrix deterministic equivalent extends both laws to finite dimension, renormalizing the anchor to $\rho_{\mathrm{eff}} = \rho (e^{\zeta q} - 1) / (\zeta q)$; it matches simulation within 5% at fixed operating points and within 13% at the method's own, and rank-deficient populations reduce exactly to their active subspace. The resulting method, SPAR, calibrates window and strength online from residuals. On a 27-configuration grid the calibrated window opens on drift without divergence, and the grid defaults transfer to an hourly electricity stream and to BCI IV 2b. The core algebra, refresh chain, and regret bounds are formalized in Lean 4.
Manaka Ueno, Geoffrey Wolfer
Tue 17 Nov
stat.MLcs.LGmath.STstat.TH
pdf html
- interest Adaptive data collection and anytime-valid inference 0.84
- your work near “Valid Inference with Imperfect Synthetic Data” 0.78
- favorites near “The Minimax Regret of Sequential Probability Assignment, Contextual Shtarkov Sums, and Contextual Normalized Maximum Likelihood” 0.84
Abstract
We initiate the study of estimating a discrete distribution with respect to the hockey-stick divergence from independent samples. We provide non-asymptotic upper and lower bounds on the minimax expected loss in terms of the alphabet size, the number of samples, and the divergence parameter. The upper bounds are achieved by add-constant estimators, while the lower bound is derived via a Bayes risk argument. Both upper bounds and lower bounds reveal a transition between polynomial and exponential regimes as the parameters of the problem vary.
Fernandes Machado Agathe, Jamotton Charlotte, Charpentier Arthur
Fri 13 Nov
stat.MLcs.LG
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.78
- interest Reinforcement learning, policy learning and off-policy evaluation 0.81
- favorites near “Preference-based Online Learning with Dueling Bandits: A Survey” 0.83
Abstract
Fairness interventions often targets disparities associated with a single sensitive attribute in binary classification tasks. However, when multiple sensitive attributes are associated with a data subject, improving fairness with respect to one attribute does not necessarily improve it across all attributes: it may exacerbate disparities associated with other sensitive attributes, including those that are unavailable to the practitioner. We refer to these unintended effects as cross-attribute externalities of fairness. We study how fairness with respect to one sensitive attribute varies when an initially unfair classification or regression model is corrected with respect to another attribute. We characterize these externalities using Wasserstein barycentric disparities and develop a distributional framework that captures changes in both marginal and joint fairness measures. This highlights a limitation of conventional fairness evaluation, in which the success of an intervention is assessed mainly for the attribute it targets. In a jointly Gaussian model, we derive explicit trajectories for marginal and joint disparities and identify conditions under which an intervention eliminates disparities with respect to its target attribute while increasing disparities associated with another attribute. We also derive sharp partial-identification bounds on the resulting externalities when a sensitive attribute is unavailable and extend our results to categorical sensitive attributes. We illustrate these externalities using classification and regression real-data: in the classification experiments, interventions increase unfairness with respect to the hidden attribute in the majority of scenarios across more than half of the datasets considered; in the regression experiments on the ACSIncome dataset, correcting for race reduces sex disparities, whereas subsequently correcting for age increases them.
Hannah Li
Fri 13 Nov
stat.MEcs.AI
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.79
- interest Reinforcement learning, policy learning and off-policy evaluation 0.81
- favorites near “Dynamically Optimal Treatment Allocation” 0.81
Abstract
In interactive systems where users engage with LLMs, users are adaptive and, after observing LLM responses, can change their prompting strategy, as well as the type and difficulty of subsequent tasks. Thus model responses can influence the set of prompts on which the model is subsequently evaluated. We study the implications of this user adaptivity and prompt endogeneity for the evaluation of LLMs. We define two classes of estimands for comparing models: the common-distribution effect, which compares model performance on the same distribution of prompts, and the induced-distribution effect, which compares each model on the prompt distribution generated through user interaction. We show that offline benchmark evaluation and live A/B testing target different quantities, with the standard difference-in-means estimators for the two methods converging to the common-distribution effect and the induced-distribution effect, respectively. This distinction can affect decision-making: the two comparisons can rank models differently and lead to discrepancies in deployment recommendations. We characterize when such ranking reversals occur and provide a sufficient condition under which the rankings are guaranteed to agree. This result provides a practical heuristic for assessing whether offline evaluation is sufficient to reproduce the decision suggested by a live experiment. Third, we show, in a stylized linear model of interactions, that user adaptivity can increase variance in estimation. Finally, we discuss implications of user adaptivity for the interpretation of experiments and show that adaptivity creates a common mechanism leading to common evaluation challenges, such as reward hacking and novelty effects. Together, this work highlights the necessity of measuring both model performance as well as changes in the user behavior that the model induces.
Kaixuan Ji, Quanquan Gu
Fri 13 Nov
cs.LGstat.ML
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.78
- reading online, sequential, far, imply
- interest Adaptive data collection and anytime-valid inference 0.83
Abstract
We study smooth calibration error in sequential binary prediction, a measure of how far probabilistic forecasts are from being calibrated that is robust to small perturbations of the forecasts. The minimax smooth calibration error is known to lie between $\Omega(T^{1/3})$ and $O(\sqrt T)$(Qiao and Zheng, 2024), where $T$ is the number of rounds. We close this gap for both adaptive and oblivious adversaries. We construct an adaptive adversary whose outcome mean moves with the forecaster's cumulative residual, which makes compensating for past residuals costly and forces every forecaster to incur $\Omega(\sqrt T)$ expected smooth calibration error. Simulating this adversary against the forecaster's average prediction yields an oblivious adversary with the same guarantee. These lower bounds match the known $O(\sqrt T)$ upper bound up to universal constants, and are therefore minimax optimal. Our results also imply matching $\Theta(\sqrt T)$ bounds for calibration distance and lower calibration distance. The proof ideas are partially developed by GPT-6 Astra under the authors' guidance.
Alokesh Manna, Debangan Dey
Mon 16 Nov
stat.MEstat.ML
pdf html
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.81
- favorites near “Priors for the Long Run” 0.81
- interest Bayesian computation and structural estimation 0.82
Abstract
Many studies record several variables at the same sites over a sequence of periods, such as pollutant concentrations at monitoring stations or case counts of infectious diseases by district. A central question concerns the dependence between the variables and its change over time. We develop dynamic bigraphical Mat\'ern--Whittle (DBMW) processes for such data. Each variable has a spatial field that evolves by autoregression. A graph over the variables couples the temporal innovations of the fields. Its links appear and vanish over time, and a link may change sign. A graph Laplacian on the sites gives the spatial structure, and a canonical exponential family links the fields to Gaussian, count or binomial responses. The joint distribution of the latent fields over all sites, variables and periods factorizes over the eigenmodes of the graph Laplacian, and its normalizing constant is free of the graph and of the dynamics. Given the fields, the posterior of the time-varying signed graph is therefore exact, by forward--backward recursions. The estimator has three steps: a fit without a graph, which also gives the switching rates with an interval, the exact graph posterior, and a joint refit. In simulations, DBMW recovers the graph with higher average precision than graphical lasso, time-varying graphical lasso and Poisson log-normal networks, and it predicts withheld data better than a standard spatio-temporal model that fits each variable separately, with a Mat\'ern field and autoregressive evolution, by integrated nested Laplace approximation. In weekly air-quality data from 288 U.S. sites, it recovers the seasonal change of sign between ozone and nitrogen dioxide. In German surveillance data on seven diseases, the district-level links stay positive and stable through the COVID-19 pandemic.
Mohamed El Omari, Saralees Nadarajah
Tue 17 Nov
math.STstat.TH
pdf html
- reading motion, consistent, convergence asymptotic, method moments
- favorites near “Priors for the Long Run” 0.81
- interest Macroeconometrics and time series 0.80
Abstract
We investigate the problem of estimation for a tempered fractional Brownian motion with drift of the form: $X_t=\mu t + \sigma B_{H,\lambda}(t)$, for all $t\geq 0$. Based on discrete observations of $X$ and by using the method of moments with the ergodic theorem, we construct estimators of the drift parameter $\mu\in\R$ and noise parameters $ \sigma^2,H,\lambda \in(0,\infty)$. We establish the asymptotic properties of the proposed estimators, including convergence in probability, almost sure convergence, and asymptotic normality. Finally, by using Gladyshev's results \cite{gladyshev1961new} we provide some consistent alternative estimators of $\sigma^2$, $H$ and $\lambda$, when $\mu=0$ and $H\in (0,1)$.
Enis Chenchene, Anil S. Baslamisli, Paolo Braca
Mon 16 Nov
stat.MLcs.LG
pdf html
- interest Kernels, operators and functional data 0.87
- favorites near “Operator-valued Kernels for Learning from Functional Response Data” 0.87
- reading empirical, linear quadratic, statistical decision, modern
Abstract
Foundation models produce high-dimensional latent representations in which suitably designed statistical decision rules can be effective yet analytically tractable. We study such detectors in the high-dimensional regime, where latent dimension and training sample size grow proportionally. For linear and quadratic statistics, we derive explicit large-deviation characterizations of false-alarm and miss-detection performance under the Neyman-Pearson paradigm, and show how the binary analysis extends to multi-class classification. The resulting error probabilities and corresponding decay rates exhibit non-monotonic dependence on the shape parameter, including double-descent and, in some regimes, multiple-descent. Experiments on vision, text, and audio datasets using modern foundation models show close agreement between theoretical predictions and empirical performance. In particular, the proposed framework allows us to set a prescribed false-alarm level and accurately predict the corresponding miss-detection probability. Overall, our results provide a theoretically grounded and interpretable framework with competitive empirical performance across different modalities.
Cosmas Kambale Musambi, Ernest Fokoué
Mon 16 Nov
stat.ME
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.81
- reading doubly robust, democratic, engines, doubly
- interest Causal inference and identification in econometrics 0.78
Abstract
Routine data on preventive chemotherapy for schistosomiasis are confounded, clustered, heavy-tailed and heaped. We study how cross-fitted augmented inverse probability weighted (AIPW) estimators behave, and how they should be computed, when least squares is replaced by Student-t outcome engines: Huber and t iteratively reweighted least squares, and a two-component t scale-mixture regression fitted by an ECM algorithm, alongside a residual-clipped augmentation. We give four short propositions (exact AIPW bias, the efficiency bound, the estimand of clipping, monotone ascent of a scale-constrained ECM), a single-file Python implementation with cluster-level cross-fitting, cluster-robust standard errors and a cluster bootstrap, and a Monte Carlo study with up to 1000 replications per cell. Robust outcome engines leave AIPW unchanged (RMSE differences at most 0.002), whereas clipping cuts RMSE by 26-45% under symmetric contamination, under-covers, and, as a plasmode experiment built on real covariates and residuals shows, shifts the estimand when residual laws differ between arms. One-sided artefacts that differ between arms bias every estimator. Heaped data also break naive code: the MAD scale collapses and the mixture likelihood is unbounded, and we give simple guards. Applied to 1485 rounds in 362 health zones of the Democratic Republic of the Congo (ESPEN, 2014-2025), integrated delivery appears to raise programme coverage by 17.4 percentage points, yet epidemiological coverage differs by only 0.9: the gap comes from targeting denominators, a real differential artefact that no outcome engine can remove.
Chenxin Lyu, Ruodu Wang
Tue 17 Nov
math.STstat.MEstat.TH
pdf html
- interest Semiparametric inference and debiased machine learning 0.86
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.83
- favorites near “Optimal Decision Rules for Weak GMM” 0.85
Abstract
We study the false discovery rate (FDR) of the Benjamini-Hochberg procedure for $K$ hypotheses at nominal level $q$ under Gaussian dependence of one-sided p-values. We ask whether the global null is least favorable after optimizing over correlation matrices. We show that the global null attains the worst-case FDR for $K\le3$ at every level. In dimension eight, an explicit mixed-null construction has FDR $21q/20$ for $0<q<1/2$. It exceeds the global-null maximum for all sufficiently small $q$, by a uniform extension of the known fixed-matrix small-level Simes bound. For $0<q<1/2$, we identify eight as the first inflation dimension in the monotone rank-one class and prove that $21q/20$ is its exact eight-dimensional maximum. In the same level range, the sharp multiplicative bound over all dimensions in this class is twice the omega constant $w = W(1)$, where $W$ is the Lambert $W$ function. This bound is approached as the dimension increases. We also give scale conditions for stability around the independence benchmark $m_0q/K$, where $m_0$ is the number of true nulls, and obtain a dimension-free upper bound of $q+q^2/2+O(q^3)$ at small levels under the global null with nonpositive correlations.
Chang Jun Im, Jeong Min Jeon
Tue 17 Nov
math.STstat.TH
pdf html
- reading wasserstein space, theory, large class, large
- interest Kernels, operators and functional data 0.82
- favorites near “Operator-valued Kernels for Learning from Functional Response Data” 0.83
Abstract
Fr\'echet regression extends regression analysis to metric space-valued responses such as probability distributions. Its asymptotic theory often relies on local entropy conditions, which control the complexity of small balls in the response space. A commonly used condition requires the associated local entropy integral to remain bounded as the balls shrink, but this requirement can fail when the response space is infinite-dimensional. We determine the exact rate at which this integral diverges in the Wasserstein space of probability distributions supported on a compact interval. Specifically, for a broad class of ball centers, we establish matching upper and lower bounds for the local covering entropy and show that the local entropy integral grows at the rate of the reciprocal square root of the ball radius. This rate persists when the response space is restricted to a large class of continuous distributions. The same rate holds for uniform local entropy integrals, with the supremum taken over ball centers. These results quantify the local complexity of the Wasserstein space and provide sharp benchmarks for entropy assumptions in statistical problems with distribution-valued responses.
Leheng Cai, Moshu Xu, Zhou Zhou
Wed 18 Nov
stat.ME
pdf html
- interest Macroeconometrics and time series 0.84
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.79
- favorites near “Spectral State Space Models” 0.76
Abstract
We study general distributional change point analysis for high-dimensional time series under nonstationary temporal dependence and limited computational resources. To mitigate noise accumulation from irrelevant coordinates and reduce the computational cost, we propose a sparse random projection procedure based on characteristic functions. We quantify the computational complexity of the random projections and show that the required number of projections grows polynomially with the ambient dimension. To accommodate nonstationary temporal dependence, a Gaussian multiplier bootstrap based on contrasts of local empirical characteristic functions is developed. We establish asymptotic false detection control and simultaneous guarantees for estimating the number and locations of multiple change points, with a near-parametric uniform localization rate up to logarithmic factors. A key technical innovation is a Gaussian approximation for high-dimensional multi-scale sparse time series, which is of independent interest. Simulation studies demonstrate the effectiveness of the proposed procedure in detecting and localizing distributional changes under various temporal dependence structures. An application to neural recordings further illustrates the practical utility of the proposed method in detecting diverse distributional changes in neural activity.
Mehrdad Pournaderi
Tue 17 Nov
math.STstat.MEstat.MLstat.TH
pdf html
- interest Online learning and statistical learning theory 0.84
- your work near “Timing as an Action: Learning When to Observe and Act” 0.77
- favorites near “Orthogonal Statistical Learning” 0.86
Abstract
Weighted split conformal prediction handles covariate shift by reweighting calibration scores with the likelihood ratio $v=dQ_X/dP_X$; its training-conditional coverage is a functional of the self-normalized weighted empirical distribution function $\hat F_m$, and we study the process $\sqrt m(\hat F_m-F_Q)$. It converges in $\ell^\infty$ to the Gaussian process $W(G(t))-F_Q(t)W(K^2)$, with $G(t)=\mathbb{E}_Q[v\mathbf{1}\{S\le t\}]$ and $K^2=1+\chi^2(Q\|P)$, if and only if $\chi^2(Q\|P)<\infty$ (for non-degenerate scores); $\sup v$ plays no role. The limit gives the joint law of the coverage across levels, consistent bootstraps, inflations asymptotically exact uniformly in the level, and Berry-Esseen-type rates. With estimated ratios, a uniform expansion remains valid when the estimate uses the calibration covariates; the estimation error enters through a covariance term that can reduce the variance below the oracle's, and a pairs bootstrap reproduces it. When the ratio is known, the weighted conformal threshold is semiparametrically efficient if and only if the ratio and the coverage indicator are uncorrelated under the test distribution, with an explicit efficient correction otherwise. When $\chi^2(Q\|P)=\infty$, the ratio has a regularly varying tail of index $p\in(1,2)$ and the score decouples from it in the far tail, the finite-dimensional limits are $p$-stable at the rate $m^{1-1/p}$ up to a slowly varying factor.
Afonso S. Bandeira, Yue M. Lu, Petar Nizić-Nikolac
Thu 19 Nov
math.PRcs.DSmath.STstat.MLstat.TH
pdf html
- interest Kernels, operators and functional data 0.85
- favorites near “Eigendecompositions of Transfer Operators in Reproducing Kernel Hilbert Spaces” 0.84
- reading random linear, matrix concentration, previously observed, results rely
Abstract
We study the spectral distribution of the least squares solution to a random linear system whose unknown is a symmetric matrix. In the normalized and isotropic setting, we derive a closed-form description of the support of the spectral distribution. As an application, we consider the ellipsoid fitting problem: given $n$ standard Gaussian vectors in $\mathbb{R}^d$, does there exist a centered ellipsoid passing through all of them? It was recently shown that this problem has a sharp phase transition at $n \approx d^2/4$. On the satisfiability side, unlike earlier results, these sharp results do not rely on least squares candidates. We show that the least squares candidates considered in earlier works have a threshold of positive semi-definiteness at $n \approx d^2/10$, which was previously observed numerically and predicted non-rigorously using statistical physics methods. Moreover, we show that the corresponding least squares dual certificates have a threshold of positive definiteness at $n \approx 2d^2/5$. Our approach combines several matrix concentration tools. A key technical ingredient, which may be of independent interest, is showing that a certain kernel matrix is well-conditioned with high probability until $n \approx d^2/2$.
Manuele Leonelli
Tue 17 Nov
stat.ME
pdf html
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.79
- reading amounts, graphs, generally, covariances
- interest Semiparametric inference and debiased machine learning 0.83
Abstract
Conditional independence in a Gaussian vector is the vanishing of almost-principal minors of its covariance matrix $\Sigma$. A sensitivity analysis replaces $\Sigma$ by the Schur product $\Delta\circ\Sigma$, where the symmetric multiplier $\Delta$ differs from one exactly at the changed positions. Such a change generally destroys some conditional independences, and we determine which minors stay zero. A vanishing minor is preserved for every matrix exactly when the corresponding block of $\Delta$ has rank one. When variances or covariances change, each almost-principal minor of $\Delta\circ\Sigma$ is a polynomial in the amounts added, with minors of $\Sigma$ as coefficients, and a conditional independence survives every such change exactly when they vanish. In an undirected graphical model, changing variances joins the neighbours of the perturbed vertices, while changing covariances leaves a structure read from the graph by a separation rule, possibly not graphical. Both extend to ancestral graphs, directed acyclic graphs included, and identify the entries that can vary freely. For the other entries, blocks of rank one along a junction tree characterize the multipliers that preserve a decomposable model. A sensitivity analysis can therefore rest on the conditional independence structure after the change, as illustrated on examination marks of 88 students.
Linglingzhi Zhu, Dongze Wu, Yao Xie
Mon 16 Nov
math.OCstat.ML
pdf html
- your work near “Efficient Online Estimation of Causal Effects by Deciding What to Observe” 0.82
- interest Semiparametric inference and debiased machine learning 0.84
- reading moment, matrices, row column, low-rank matrices
Abstract
We study distributionally robust risk under perturbations restricted to low-rank matrices whose row and column spaces are shared across observations, focusing on how much worst-case risk this restriction sacrifices and how to compute the restricted worst case. Under moment or Wasserstein budgets, we derive quantitative bounds on the gap between the unrestricted and low-rank worst-case risks, characterize when the gap vanishes, and identify the smallest lossless row and column dimensions. To generate worst-case perturbations, we apply projected gradient ascent to a curvature-corrected concave problem and derive computable risk certificates. We establish global sublinear convergence for the corrected objective and linear convergence to a worst-case distribution when the budget constraint is sufficiently active.
Cole Becker, Taylan Kargin, Kimia Hassibi, Asuman Ozdaglar, Pablo A. Parrilo
Mon 16 Nov
stat.MLcs.AIcs.LG
pdf html
- favorites near “Monte Carlo Geometry Processing” 0.82
- reading rational, good, discretization, functions
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.78
Abstract
Generative modeling based on stochastic interpolants requires specifying two key choices: an interpolant schedule bridging the initial and target distributions, and an efficient discretization of the resulting probability-flow ODE. In this work, we introduce a new framework for joint schedule/discretization design based on the analyticity properties of the (complexified) ODE trajectories, which in turn ensure good numerical behavior. Our formulation can be solved exactly via conformal mapping, yielding a natural family of variance-preserving elliptic schedules, parametrized by Jacobi elliptic functions. Pairing an elliptic schedule with a tailored discretization, we obtain Zoloflow, a new sampling algorithm closely connected to Zolotarev's optimal rational function approximants. Crucially, both the elliptic schedule and Zoloflow explicitly incorporate the condition number $\kappa$ of the target distribution and reduce to the classical trig schedule in the limit $\kappa \rightarrow 1$. For Gaussian targets, Zoloflow achieves minimax optimality among all oblivious score-span algorithms, attaining total-variation error $\varepsilon$ using $\Theta(\log\kappa\,\log(\sqrt d/\varepsilon))$ (exact) score queries. For the log-concave case, the elliptic schedule allows sampling with at most $\varepsilon$ 2-Wasserstein-error by requiring merely $\widetilde{O}(\sqrt{d} \, \mathrm{polylog}(\kappa, 1/\varepsilon))$ score evaluations. Complementing these theoretical results, we empirically evaluate the elliptic schedule on general (non-log-concave) target distributions, including detailed comparisons with alternative schedule proposals on both synthetic benchmarks and existing image generation datasets.
Saba Esnaashari, Tvesha Sippy, Youmna Hashem, Sukankana Chakraborty, Jonathan Bright
Fri 13 Nov
stat.AP
pdf html
- reading productivity, technology, south, questions
- favorites near “When Did Growth Begin? New Estimates of Productivity Growth in England from 1250 to 1870” 0.77
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.70
Abstract
Artificial intelligence has the potential to transform how public services are delivered, with important implications for public service productivity. But how widely is the technology actually being used - and is it making any difference? Our survey of 3,362 public sector workers across the UK, the USA, South Africa and the EU finds widespread use of AI in every region. Our findings suggest that this spread is being driven by bottom-up individual adoption rather than top-down institutional adoption. In the EU, for example, 65% of respondents use AI, while only 19% say their organisation has formally adopted it. Most users are positive about its effects, with 79% saying AI has improved their productivity. However, only 29% say it has increased what is expected of them at work, raising questions about how truly transformative the technology is so far. Across regions, South Africa is a clear outlier, with the highest levels of use, organisational adoption, optimism and awareness of guidelines.
Yibo Wang, Chenlei Leng, Guosheng Yin
Tue 17 Nov
stat.ME
pdf html
- your work near “Automated Solution of Heterogeneous Agent Models” 0.83
- interest Bayesian computation and structural estimation 0.83
- favorites near “Empirical Bayes for the Reluctant Frequentist” 0.87
Abstract
We develop scalable estimation and inference for directed probit networks with correlated sending and receiving effects and residual reciprocity. An incomplete composite likelihood in two stages combines exact bivariate margins of sampled dyads and triples, avoiding integration in high dimensions and cubic enumeration. For fixed parameter dimension, criterion evaluation is linear in the number of draws. Averaging a fixed number of fits from independent index samples reduces sampling variation and allows parallel computation. We establish consistency and joint asymptotic normality with different directional rates, allowing a singular node projection when draw counts lie between the node and dyad scales. An overlap covariance accounts for network and index variation and is consistent relative to each directional variance. Independent index samples stabilize covariance estimation without refitting. The resulting procedure provides standard errors, mapped joint regions and score tests for zero variance components. Simulations show improved estimation from averaging and substantially lower computation times than default Bayesian fitting. An application to a fly brain network with 21,738 neurons demonstrates estimation and inference at a scale relevant to modern network data.
Hirofumi Ota
Fri 13 Nov
math.PRmath.STstat.TH
pdf html
- reading processes, gaussian process, uniformly, continuous
- favorites near “Bernstein-von Mises theorems for time evolution equations” 0.87
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.78
Abstract
For a centered Gaussian process $G$ with continuous sample paths on a compact metric space $T$, a continuous drift $f: T \mapsto \mathbb{R}$, a noise level $\sigma>0$, and $\epsilon\ge 0$, let $A_\epsilon$ be the set of points at which $f+\sigma G$ is within $\epsilon$ of its maximum. Writing $\mathbb{E}\sup_TG$ for the expected supremum of $G$, we prove that the expected squared diameter of $A_\epsilon$ in the canonical pseudometric of $G$ is at most $8\epsilon \mathbb{E}\sup_TG/\sigma$ and that the expected maximal squared distance from a point of $A_\epsilon$ to an exact maximizer is at most $4\epsilon \mathbb{E}\sup_TG/\sigma$. The constants cannot be improved uniformly over index sets, processes, and drifts, and no logarithmic factor in $\epsilon$ appears, unlike in bounds integrated from tail estimates. At $\epsilon=0$ the bounds give the almost sure uniqueness of the maximizer in the canonical pseudometric without nondegeneracy assumptions. Given this uniqueness, both bounds follow from a bound with constant $2$ on the mean square of the largest increment of an independent copy of $G$ from an exact maximizer to a point of $A_\epsilon$. The proof rests on an identity for the mean squared support function of the $\epsilon$-subdifferentials of a smooth convex function at a standard Gaussian vector, obtained from two Gaussian integrations by parts.
Pranav Gupta, Satyaki Mukherjee
Tue 17 Nov
math.STmath.OCstat.COstat.MLstat.TH
pdf html
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.79
- favorites near “Geometric MCMC for Infinite-Dimensional Inverse Problems” 0.84
- reading nonparametric, en, higher, nondegenerate
Abstract
We study the computation of approximate nonparametric maximum likelihood estimators (NPMLEs) for Gaussian location mixtures in arbitrary dimension $d\ge 1$ via the Fisher-Rao gradient flow. We provide a certified algorithm based on an entropic proximal discretization of the flow which, given a sufficiently small $\varepsilon > 0$, returns an absolutely continuous mixing measure $\tilde\mu$ admitting an exact finite-dimensional representation and satisfying $\hat L_n-L_n(f_{\tilde\mu})\le\varepsilon$. For almost every dataset $X_1,\cdots, X_n \in \mathbb{R}^d$ satisfying $n>2(d+1)(m+1)^2$, where $m$ is the number of support points of the NPMLE, the algorithm requires $O(n\varepsilon^{-1}\log^{d+2}(1/\varepsilon))$ time. En route, we also establish results on the well-posedness and convergence properties of the continuous-time Fisher--Rao flow. In particular, initializing the flow at the uniform probability measure on a ball containing the convex hull of the observations, we show that $\hat L_n-L_n(f_{\mu_t})=O(\log t/t)$. We also establish that, for almost every dataset satisfying the same condition on $n$ as above, the set of global maxima of the optimal score is the same as the support of the NPMLE, and the Hessian of the optimal score is nondegenerate at these points, extending a result of (Polyanskiy-Sellke, 2025) to higher dimensions.
Nong Minh Hieu, Antoine Ledent
Tue 17 Nov
stat.MLcs.LG
pdf html
- reading generalization, theory, empirical, explains
- favorites near “PAC-Bayes Compression Bounds So Tight That They Can Explain Generalization” 0.85
- interest Kernels, operators and functional data 0.80
Abstract
Empirical evidence suggests that large neural networks rarely make effective use of all available parameters, with learned solutions exhibiting pronounced sparsity. Despite this ubiquity, existing generalization theory only partially explains how such sparsity influences statistical performance. In this paper, we derive non-asymptotic generalization gap bounds for deep neural networks whose layer-wise weight matrices are restricted by $\ell_p$ quasi-norm ($0<p\leq 1$) constraints. Our analysis gives rise to a novel result that interpolates between classical norm-based generalization guarantees and parameter-counting bounds. Remarkably, in the extreme sparsity limit as the quasi-norm order $p$ approaches zero, our analysis reveals a transition to a square-root dependence on the number of \emph{active} (non-zero) parameters, resulting in bounds that are strictly tighter than existing parameter-counting approaches. Experiments demonstrate that our bounds exhibit a milder dependence on initial architectural choice at the over-parametrization regime, effectively capturing the generalization benefit of the sparsity arising from training on natural data.
Furkan Danisman, Luke Zhang, Zhijing Jin, Dehan Kong
Tue 17 Nov
stat.MLcs.AIcs.LG
pdf html
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.83
- reading counterfactual, factual, causal, treatment
- interest Language models as statistical objects 0.77
Abstract
Estimating treatment effects is central to science across many disciplines and requires counterfactual reasoning about alternative worlds. In the real world, however, each unit reveals its outcome only under the factual world. Thus, the factual and counterfactual outcomes can never be jointly learned by any model trained on real data. The problem is therefore tractable only in synthetic settings, where both worlds can be generated. Recent progress on prior-data fitted networks (PFNs) has shown that models pre-trained purely on synthetic data can transfer to real-world prediction. This paradigm has since carried over to causal effect estimation. Despite this fully synthetic setup, existing causal PFNs carry the real-world constraint into their training objective, confining learning to a single world's outcome per unit. The dependence between the two worlds is thereby lost, and the density of the treatment effect relies by default on independence, which is implausible in practice. We therefore propose joint learning, a training setup in which a causal PFN learns the joint distribution of the factual and counterfactual outcomes of each unit. From the joint distribution, we obtain both the response and the individual treatment effect densities, and at inference, we apply optimal transport across units to obtain an approximation of the average treatment effect density. Ours is therefore the first causal PFN approach to provide all three densities together while accounting for the cross-world dependence. We apply joint learning and optimal transport to existing causal PFNs and evaluate it across semi-synthetic, synthetic, and real benchmarks spanning diverse causal scenarios. The jointly trained models significantly improve upon their baselines in point estimation and uncertainty quantification in most settings, indicating that joint learning is a preferable training setup for causal PFNs.
Fabrizio Maturo, Annamaria Porreca
Fri 13 Nov
stat.MLcs.LGstat.ME
pdf html
- favorites near “Distributional Random Forests: Heterogeneity Adjustment and Multivariate Distributional Regression” 0.89
- reading functional, tuned, adjustment, failed
- interest Language models as statistical objects 0.76
Abstract
Functional random forests built on a single functional principal component decomposition of the whole domain may dilute features confined to short intervals, whose location is rarely known in advance. We propose two random forests based on window-specific functional principal components. The global-local functional random forest (GL-FRF) is a late-fusion method that stacks a tuned full-curve forest and one forest per window with penalised convex weights; it contains the full-curve forest as a feasible solution, and its weights indicate which parts of the domain drive the prediction. The window-concatenated functional random forest (WC-FRF) is an early-fusion method that trains a single tuned forest on the concatenated window-specific scores. On twenty UCR datasets, GL-FRF improved on a tuned full-curve forest with the same number of trees on 16, with Brier score reductions of up to 0.21, significantly after Holm adjustment. No method was best on more than six datasets. WC-FRF was more accurate than GL-FRF on 13 datasets but failed when the signal required the full curve, whereas GL-FRF had the smallest mean and maximum excess over the best method across all 32 simulated and real settings, which makes it the safer choice when the location of the signal is unknown.
Filip Obradović, Gabriel Ziegler
Tue 17 Nov
econ.THecon.EMstat.AP
pdf html
- interest Reinforcement learning, policy learning and off-policy evaluation 0.82
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.79
- reading viewed, debate, consistent, belief
Abstract
Dilation (Seidenfeld and Wasserman 1993) occurs when conditioning on new information \textit{only} increases uncertainty regardless of what the information reveals. Often viewed as an undesirable feature of models of ambiguity, dilation has sparked sustained debate about belief updating. This paper shows that following established clinical guidance can induce dilation, even for diagnostic tests that meet conventional performance benchmarks. We characterize when a test is dilation-inducing and show that a ubiquitous feature of performance studies -- evaluating tests against an imperfect reference -- creates scope for dilation. To facilitate detection before clinical use and inform regulatory standards, we develop a finite-sample valid hypothesis test of the null that a diagnostic test is dilation-inducing in this context. Applying the method to an influential study that preceded and contributed to the clinical adoption of chest CT screening for COVID-19, we find that the data are consistent with dilation.
Lea L. Mairhöfer, Jonathan B. Krause, Björn-Hergen Laabs, Silke Szymczak
Tue 17 Nov
stat.MEcs.LGstat.ML
pdf html
- favorites near “Distributional Random Forests: Heterogeneity Adjustment and Multivariate Distributional Regression” 0.90
- interest Online learning and statistical learning theory 0.83
- your work near “Valid Inference with Imperfect Synthetic Data” 0.78
Abstract
Random forests (RFs) often achieve high prediction performance, but their ensemble structure can be difficult to interpret. Surrogate models and rule extraction methods offer interpretable approximations, yet guidance on selecting an appropriate method for a specific research question remains limited. We compared 12 methods for interpreting RFs, with a decision tree as a prediction-oriented reference, across 11 simulation settings and 67 benchmark datasets for regression and classification. Evaluation addressed prediction and structural similarity to the RF, model complexity, prediction performance, and runtime. Simulations additionally assessed explainability using known effect and noise variables. No method performed best across all objectives. For regression, SLIM provided the strongest general performance for prediction similarity or performance on test data, while a linear model offered a simpler global interpretation of linear effects when performance was comparable. For classification, inTrees achieved the strongest overall benchmark performance, with CTree and OC-CART providing useful alternatives. ART consistently produced the most compact representations. NH and SIRUS offered strong structural similarity for regression, while MRT most closely reproduced RF variable usage in the classification simulations, illustrating that prediction-oriented rankings do not necessarily identify the best structural explanation. Outcome-type restrictions, runtime limits, and method-specific failures reduced result availability for several approaches and should be considered alongside performance when selecting a method. Method selection should therefore be guided by the outcome type, anticipated data structure, and intended interpretability objective. Prediction similarity alone is insufficient for explaining an RF and should be considered jointly with structural similarity and model complexity.
Karl Friedrich Siburg
Tue 17 Nov
math.STstat.TH
pdf html
- interest Semiparametric inference and debiased machine learning 0.84
- reading little, empirical, spectrum, dependence measures
- your work near “Perturbation Methods for Incomplete Markets Economies” 0.80
Abstract
For a bivariate copula $C$ with $(U,V)\sim C$, the conditional distribution functions $L_t=\partial_1C(U,t)$ are random variables with means $t$, so that measures of directed dependence become risk functionals. The absolute Lorenz curves of the variables $L_t$ form the rearranged copula $C^\uparrow$. In this picture Chatterjee's rank correlation $\xi$ is a squared Sobolev seminorm and the rearranged Spearman correlation an $L^1$ norm of $C^\uparrow-\Pi$, and both arise from one spectral risk measure, applied either to the conditional laws of $V$ or to the laws of $L_t$. The resulting classes $\omega_\phi$ and $\kappa_\phi$ of spectral dependence measures have mirror-image axiomatic properties. For every spectrum we determine the exact upper bound of $\kappa_\phi$ in terms of $\xi$, and we prove sharp inequalities between $\xi$, $\kappa_\phi$ and the values of $C^\uparrow$, which show how little global measures say about the tails. For plug-in estimators based on the empirical checkerboard copula we obtain consistency, rates and a resolution limit, and a simulation study shows that tail-localised measures separate copulas with identical $\xi$. An application to large credit portfolios illustrates the results.
Terence Viaud, Ioannis Kontoyiannis
Tue 17 Nov
cs.ITmath.ITmath.STstat.TH
pdf html
- reading science, computer science, partly, emphasizing
- interest Kernels, operators and functional data 0.80
- favorites near “Introduction to regularity structures” 0.82
Abstract
We introduce a new framework for the study and evaluation of the fundamental limits of channel coding, emphasizing genuinely nonasymptotic results. The {\em channel coding sample complexity} $n^*({\cal W},\epsilon)$ of a channel ${\cal W}$ is defined as the smallest blocklength at which it is possible to transmit information through that channel with error probability below $\epsilon$ and at a rate at least as high as that given by an appropriate constraint also in terms of $\epsilon$. This formulation follows analogous developments in statistics, computer science and, more recently, in lossless source coding. For any discrete memoryless channel we derive nonasymptotic bounds on the sample complexity, with explicit constants. These bounds show that the sample complexity is characterized not by the usual channel capacity, but by the channel's Augustin--Csisz\'{a}r capacity of order $1/2$, which coincides with the cutoff rate as well as with the channel's Sibson capacity of order $1/2$. The proof is partly based on a new channel-coding one-shot converse in terms of hypothesis testing. We also examine extensions of the sample complexity framework to general target rates. Moreover, we derive corresponding results for the sample complexity of the additive white Gaussian noise channel with a quadratic power constraint. Their proof relies on new, explicit upper bounds on the best achievable error probability, which may be of independent interest.
Anton Badev
Tue 17 Nov
econ.EMcs.SI
pdf html
- reading economists, theory, policy, appear
- favorites near “Labor Rationing” 0.77
- interest Reinforcement learning, policy learning and off-policy evaluation 0.77
Abstract
Teams emerge organically when individuals decide on their own their activities and collaborators. I develop and test a theory in which teams form with consent, interactions reflect team composition, and payoffs feature externalities and several forms of congestion. Stability constraints concern group interactions and require each team member to be willing to remain rather than withdraw or pursue a project alone. Empirical analysis of research collaborations among economists at the Federal Reserve Board suggests that: (i.) these constraints help explain observed teams beyond the payoffs to collaboration; (ii.) estimated violations point to large teams of entrants that appear to be assigned; (iii.) a compensatory mechanism mitigates the effects of variation in policy workload, which appears to affect team composition more than total research output; and (iv.) shifting workload toward experienced economists increases entrants' research output and collaboration, at the cost of lower output among experienced economists who compensate for this shift in policy workload with increased interactions with entrants.
Simon Vary, Alireza Mousavi-Hosseini, Roger Grosse
Tue 17 Nov
cs.LGstat.MEstat.ML
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.81
- favorites near “Using Imperfect Surrogates for Downstream Inference: Design-based Supervised Learning for Social Science Applications of Large Language Models” 0.84
- interest Language models as statistical objects 0.79
Abstract
Causal validation of training data attribution (TDA) methods is costly, since evaluation involves retraining a model on many different subsets of the original training set. For example, the Linear Datamodeling Score (LDS), which has become a standard metric for evaluating TDA methods, measures the correlation between TDA predictions and expected retrained model outputs. In this paper, we reduce the amount of computation needed to estimate the LDS along three axes: the number of sampled training subsets, the number of retraining seeds per subset, and the number of models in the attribution ensemble. Each of these factors induces a form of bias in the LDS estimation, and removing these sources of bias significantly reduces the number of retrainings needed to match the same accuracy in LDS estimation. On ResNet-9/CIFAR-10, finite-subset debiasing achieves $3.2\%$ relative RMSE with as few as $4$ subsets, and $32$ subsets match the accuracy of ordinary Spearman using $96$ subsets, measured against a 300-subset reference. On ResNet-9/CIFAR-10, BERT/QNLI and GPT-2/WikiText-2, estimating the training-noise variance allows us to reproduce the standard five-seed LDS at the same accuracy with $2$--$5\times$ fewer retrainings.
Ali Amini, Mohammad Alijanpour Shalmani, Kaare B Mikkelsen, Robertas Damaševičius
Fri 13 Nov
stat.AP
pdf html
- favorites near “Spectral State Space Models” 0.81
- reading forecasting, consistent, references, online
- your work near “Timing as an Action: Learning When to Observe and Act” 0.74
Abstract
Sleep is scored afterwards in 30-second epochs as wake (W), the transitional stage N1, or N2. We study EEG forecasts, issued every 30 s from the preceding 30 s of EEG alone, of whether the first scored N2 epoch will begin within five minutes, on Sleep-EDF. A Brier score pooled over decisions made before and after the input is scored N1 does not show when the gain arose. In our own prespecified one-shot held-out comparisons, the spectral model SpecLR-18 improved on a training prevalence constant by 18.3% in the internal test, whereas its prespecified external result was inconclusive. Joining the retrospectively scored stage of each input epoch to the same frozen predictions, almost all of that internal-test Brier difference was attributed to N1-scored decisions; a diagnostic reference told only the scored stage reached 26.3% in the internal test and 29.5% in the external cohort, above the spectral forecasts in point estimate though not every difference excluded zero; and within-stage EEG advantages seen in single cohorts were not consistent across cohorts, with no consistent wake-stratum improvement. These references use labels unavailable online and cannot reveal what a model represents; they show why pooled forecasting scores need state-conditioned interpretation before being read as early anticipation.
Aaditya Ramdas
Tue 17 Nov
stat.MEmath.STstat.TH
pdf html
- reading control, procedures, gives, allocations
- interest Semiparametric inference and debiased machine learning 0.85
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.78
- looks like clinical and biomedical studies
Abstract
When a discovery depends on analysis choices, a list of selected hypotheses that controls the false discovery rate (FDR) gives little indication of how easily the finding could change: the selected set may be very sensitive to a variety of analyst choices, like the target FDR level, which machine learning algorithms are used in the pipeline, the choice of test statistics, and so on. We propose that scientists should measure and control a new metric called the \emph{stable FDR} (sFDR), and give methods to control it, including simple wrappers around existing FDR analysis pipelines. sFDR is a natural continuous extension of the FDR: we assign each hypothesis an allocation between zero and one, and the sFDR equals the expected fraction of null allocation to total allocation. If one runs multiple analyses of the same data by varying different reasonable choices, allocations can be determined by simply dividing each run's selection indicators by its discovery count, then averaging across analyses and rescaling. The resulting sFDR equals the average of the ordinary FDRs of the runs, even though the runs are dependent: thus sFDR control holds under the same assumptions that the original FDR control held. (In contrast, naively reporting just fractions of times each variable was selected need not control the sFDR.) It is interpretable: hypotheses that were always selected are allocated 1 and those never selected are allocated 0. In leukemia expression and neuroimaging analyses, findings with larger allocations were more often supported in separate samples. We also develop procedures that can assign stable allocations directly from p-values, e-values, or knockoff statistics, resulting in stable-BH, stable e-BH and stable knockoff procedures. A representation by e-values describes which allocation rules control the error rate and gives ways to optimize allocations for different follow-up goals.
Vincent C. Scholz, Phaedon-Stelios Koutsourelakis
Mon 16 Nov
cs.CEcs.LGstat.ML
pdf html
- favorites near “Neural Operator: Learning Maps Between Function Spaces With Applications to PDEs” 0.82
- reading wrong, keeping, functional, functional form
- interest Kernels, operators and functional data 0.80
Abstract
The hyperelastic law of a material such as soft tissue differs between specimens, and often only a few specimens can be tested. Physics-augmented neural networks (PANNs) learn such laws from data without a prescribed functional form while keeping them physically admissible, and Bayesian versions of related models quantify the uncertainty of the law of one specimen or of a pooled group of specimens. A pooled fit, however, attributes the differences between specimens to measurement noise or to uncertainty about a mean law, and neither describes a new specimen. We first develop a Bayesian PANN for one specimen, whose priors keep the learned law plausible for few, noisy data. We then extend it to a hierarchical Bayesian PANN, a nonlinear mixed-effects model in which all specimens share learned polyconvex features and each specimen is a draw from a population of laws. Features, specimens, and population are inferred jointly by variational inference, with a separate model of the measurement noise, so that the variability between specimens is not mistaken for noise, the population carries its own uncertainty, and every specimen drawn from it is polyconvex. The population predicts the response of new specimens with credible intervals, without data or from a partial test, and assigns them to cohorts with quantified uncertainty, all without a prescribed form of the law. On synthetic populations, it recovers the population from few specimens and predicts new specimens more reliably than three published methods. On Treloar's rubber and on healthy and aneurysmal aortas, it reaches the accuracy of a deterministic PANN and assigns no aorta specimen to the wrong cohort with confidence, although it underestimates the uncertainty of untested curves.
Jacob A. Johnson, Felix Jimenez, Matthias Katzfuss
Thu 19 Nov
stat.MEstat.AP
pdf html
- reading mcmc, empirical, science, complicated
- favorites near “Priors for the Long Run” 0.82
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.77
Abstract
Transport maps provide a powerful framework for the generative modeling of complex, non-Gaussian spatial fields, such as those encountered in climate science. By mapping a complicated target distribution to a standard Gaussian reference distribution, they can serve as efficient stochastic surrogates. However, current Bayesian transport maps, which are modeled using autoregressive Gaussian processes, typically rely on empirical Bayes estimates that ignore hyperparameter uncertainty. Here, we contrast this point-estimate approach with a fully Bayesian model that provides posterior uncertainty quantification in transport map hyperparameters. To avoid the high cost of Markov chain Monte Carlo (MCMC), we also propose a Laplace approximation to the hyperparameter posteriors, which can be computed at negligible additional computational cost relative to empirical Bayes. We provide numerical evidence and a real-data climate application demonstrating that these Laplace approximations can accurately capture most posterior summaries at a fraction of the cost of MCMC, while maintaining good predictive properties. Thus, our approach offers a scalable, uncertainty-aware framework for the generative modeling of non-Gaussian climate fields.
Cinzia Di Nuzzo, Matteo Farnè
Fri 13 Nov
stat.MEstat.AP
pdf html
- reading bank, high-dimensional data, make, matrices
- interest Semiparametric inference and debiased machine learning 0.83
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.78
Abstract
Factorial and Reduced K-means estimate a partition and a low-dimensional clustering subspace jointly, which makes them attractive when the variables that separate groups are not those that dominate total variance. On contaminated high-dimensional data they face three linked problems: trimming can make the scatter matrices singular, the criterion has many local optima, and the rank, number of groups and trimming proportion are unknown and interact. We propose REG-HD-TFKM and REG-HD-TRKM, regularized trimmed versions of the two methods, together with a selection procedure that handles these problems as separate decisions. A ridge-stabilized spectral step keeps every fit well defined; heterogeneous multistart search explores alternative solutions; method-specific agreement and reproducibility scores choose among them; and trimming is applied only when a radial diagnostic or the path over trimming levels supports it. A simulation study with 41 design cells shows why the separation matters. Two representations of euro-area bank balance sheets show that similar partitions can rest on nearly orthogonal subspaces, and that the same selected structure can describe different retained samples.
Blair Robertson, Chris Price, Marco Reale, Philip Davies
Fri 13 Nov
stat.MEmath.STstat.TH
pdf html
- interest Adaptive data collection and anytime-valid inference 0.85
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.82
- favorites near “Orthogonal Statistical Learning” 0.86
Abstract
Auxiliary information can improve estimation precision under probability sampling, but the relationship between the study and auxiliary variables may not be known well enough to specify suitable balancing variables. Distributionally balanced designs address this by targeting agreement between sample and population auxiliary distributions. We extend energy-based distributional balance to prescribed unequal inclusion probabilities by comparing the empirical sample distribution with an inclusion-probability-weighted population distribution. Under a regularity condition relating the study values divided by their inclusion probabilities to the auxiliaries, we establish an upper bound on the design variance of the Horvitz-Thompson estimator in terms of the expected energy distance. This bound motivates an algorithm that exchanges units between randomly paired candidate samples without increasing their average energy distance. Combined with random initialization and uniform selection of a final candidate, the procedure preserves the prescribed first-order inclusion probabilities. We evaluate its auxiliary-space properties and estimation performance using generated populations and the Meuse data.
William J. Dwyer, David Zurakowski
Tue 17 Nov
stat.ME
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.86
- reading chance, units, imply, design stage
- interest Causal inference and identification in econometrics 0.79
Abstract
Randomization makes the unadjusted treatment-effect estimate unbiased over the allocations a scheme could produce, although it does not balance any single trial. Chance imbalances on strongly prognostic baseline variables directly threaten the validity and regulatory credibility of trial results, yet what design-stage balance $P$ values imply for bias and error rates is seldom stated. We review evidence on chance covariate imbalance and derive that link under a linear model with one normal prognostic covariate, with simulation checks. Given the realized allocation, the unadjusted estimate is biased by $\beta$ times the difference in covariate means; in standard-error units the bias is $\rho Z_{\mathrm{bal}}$, where $\rho$ is the covariate-outcome correlation and $Z_{\mathrm{bal}}$ the balance z-statistic, so it does not shrink as trials grow. Pre-specified ANCOVA removes this bias, change scores reverse it, and non-collapsibility attenuates odds and hazard ratios even under perfect balance. A worst-balance threshold of $P = 1/e$ marks where the Sellke-Bayarri-Berger bound reaches 1; under complete randomization the worst of $k$ independent balance $P$ values falls below it with probability $1-(1-1/e)^k$, 0.99 for $k=10$. Any balance $P$ value $\geq 0.317$ keeps the two-sided conditional type I error of the unadjusted test at or below 5% for every $\rho$; at $1/e$ the one-sided error in the favoured direction is at most 4.1%, against 13-33% at $P = 0.05$. Design-stage balance scores should be benchmarked against complete randomization, weighted by external prognostic strength, and reported alongside allocation predictability and a pre-specified covariate-adjusted analysis plan.
Emily L. Norton, Peter Kraft, Kristen D. Brantley, Thomas Ahearn, Montserrat Garcia-Closas, James M. Hodge, et al.
Tue 17 Nov
stat.APstat.ME
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.81
- interest Semiparametric inference and debiased machine learning 0.83
- favorites near “Using Imperfect Surrogates for Downstream Inference: Design-based Supervised Learning for Social Science Applications of Large Language Models” 0.83
Abstract
Meta-analysis is foundational to federated learning, where one learns parameters by combining information across studies without sharing individual-level data. Handling missing and disparate covariates across studies is a major issue due to potential introduction of bias and decreased efficiency. We previously proposed a generalized meta-analysis estimator (GENMETA) to build a unified regression model using information on parameter estimates from sub-models and a reference population to estimate the joint covariate distribution. Building on that work, we present a linear estimator (GENMETA-Lite) as a practical, non-iterative method for multivariate meta-analysis, extended to handle heterogeneity in covariate distributions across studies, coarsened covariates, and model selection. Simulations demonstrate that our estimator has robust performance compared to estimates from na\"ive inverse-variance weighted meta-analysis or multiple imputation with individual-level data (approximately 4 to 100-fold reduction in mean squared error), even when underlying assumptions of linearity and homogeneous distributions are violated. We apply the method to fit a breast cancer risk prediction model with data from five studies (N = 313,851). Our study shows GENMETA-Lite is a robust federated learning approach for building regression models by combining information across disparate studies. An R package implementing these methods is made publicly available.
Sattwik Ghosal, Xuran Meng, Yuming Sun, Yi Li
Wed 18 Nov
stat.ME
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.79
- favorites near “Deep Learning for Individual Heterogeneity” 0.82
- reading treatment, theory, nuisance, nonparametric
Abstract
Conditional average treatment effects (CATEs) provide a principled framework for assessing treatment-effect heterogeneity and identifying patients who may benefit from an intervention. For survival outcomes, CATEs may also evolve over time and pose challenges for estimation and inference. Existing methods can accommodate complex covariate heterogeneity but often fail to exploit shared temporal structure or provide valid inference for time-varying treatment effects and their contrasts. We propose a deep survival learning framework that combines neural networks for nonparametric covariate effects with a shared spline representation of temporal variation, enabling joint estimation of CATE trajectories. Using doubly robust pseudo-outcomes and sample splitting, we establish nonasymptotic oracle bounds accounting for network and temporal approximation, nuisance estimation, and optimization errors. Our theory shows that exploiting shared temporal structure yields sharper error bounds for temporal contrasts than those derived from separate time-specific estimators. For inference, we develop a subsampled ensemble procedure and establish asymptotic normality for pointwise CATEs, temporal contrasts, and covariate contrasts, including temporal contrasts over shrinking time intervals. Under suitable regularity and bias-control conditions, consistent infinitesimal jackknife variance estimators yield asymptotically valid confidence intervals. Simulations evaluate estimation accuracy, interval coverage, and sensitivity to temporal dimension and basis mismatch. An application to lung cancer survival data illustrates the framework's ability to characterize heterogeneous treatment benefits over time.
Mansoor Ahmed, Lilia Chebbah, Jenny Dinh, Murray Patterson, Hemanth Venkateswara
Tue 17 Nov
stat.MLcs.LG
pdf html
- reading special, state, keeping, positions
- favorites near “Sequential Monte Carlo Steering of Large Language Models using Probabilistic Programs” 0.78
- interest Language models as statistical objects 0.79
Abstract
Masked diffusion models generate a sequence by revealing masked positions over many denoising steps. A few-step sampler has to reveal many positions at once, but the standard reverse kernel draws them independently and primarily ignores the conditional dependence between tokens revealed in the same step. We propose a novel framework called Domino, which models this dependence with a first-order Markov chain over the revealed positions, given the current state. The first revealed position keeps the posterior of a pre-trained masked diffusion model (MDM), and each subsequent position follows a transition given the token of its predecessor. Notably, the resulting step is normalized by construction, contains the factorized step in MDMs as a special case, and has a likelihood in closed form. We implement the Domino transitions via a lightweight module on the hidden states of a pre-trained MDM denoiser with only 2.6% parameter overhead and train by keeping the MDM denoiser frozen. Our extensive evaluations across text, molecules, and images show that Domino lowers the few-step generative perplexity of MDMs by a factor of 1.7 to 3.8 on One Billion Word (LM1B), WikiText-103, and OpenWebText at nearly unchanged sample entropy. It also raises the validity of few-step molecule samples by 1.5x. The source code is available at: https://github.com/mansoor181/domino.git
Christopher Liaw, Aranyak Mehta, Grigoris Velegkas, Felix Zhou
Fri 13 Nov
cs.DScs.AIcs.LGstat.ML
pdf html
- favorites near “Sequential Monte Carlo Steering of Large Language Models using Probabilistic Programs” 0.81
- interest Language models as statistical objects 0.81
- reading differentiation, language model, language, taylor
Abstract
Context attribution asks how removing parts of a prompt changes a language model's output. We study leave-one-out (LOO) attribution for a prompt of $n$ tokens and a fixed measurement function. Computing all LOO values directly requires $n$ model evaluations. Under a stability assumption, our algorithm estimates all LOO values to error proportional to $\varepsilon$ with constant probability using $O(\sqrt{\log(1/\varepsilon)}/\varepsilon)$ evaluation calls. To complement our upper bound, we prove a matching $\Omega(\sqrt{\log(1/\varepsilon)}/\varepsilon)$ lower bound on the number of evaluation calls in an oracle model. Empirically, we verify coefficient decay consistent with stability and evaluate the performance of our sketching algorithm on a small language-model retrieval task. Technically, our result is based on a sketching algorithm for arithmetic circuits whose Taylor coefficients satisfy the stability condition recently introduced by Gunn [Gun26]. A direct adaptation of their forward differentiation sketches would require $\widetilde{O}(\log(n)/\varepsilon^2)$ evaluation calls. Our key observation is that we can apply forward differentiation to the backpropagation circuit, with only constant overhead over a forward sketch of the same Taylor degree. This reduces the variance of the sketches and is key to our improved evaluation bound. Our lower bound builds on the techniques of Amsel, Chen, Duman Keles, Halikias, Musco, and Musco [ACDH+26, SIAM Journal on Matrix Analysis and Applications] for lower bounding the number of matrix-vector queries to estimate the diagonal elements of random matrices.
Protiva Sadhukhan Prova, Nazmul Haque, Md Asif Raihan, Md. Hadiuzzaman
Tue 17 Nov
stat.AP
pdf html
- interest Bayesian computation and structural estimation 0.84
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.79
- reading mcmc, bias-variance, sequential, empirical
Abstract
Accurate calibration of traffic speed-density fundamental diagram models is essential for freeway traffic flow analysis, yet existing approaches estimate parameters from fixed datasets without examining how estimates evolve as data availability increases. This study develops a comprehensive Bayesian stochastic calibration framework integrating sequential Markov Chain Monte Carlo (MCMC) estimation via a Metropolis-Hastings within Gibbs sampling scheme with a novel Bias-Variance Cone analysis for asymptotic parameter identification. The framework is applied to 17 macroscopic speed-density models using one full year of 5-minute PeMS loop detector data from three California freeway stations. Sequential MCMC progressively increases the number of randomly sampled days from 1 to 365, enabling systematic evaluation of parameter stability and convergence as a function of data availability. Among the 17 models, nine achieved stable posterior convergence with physically interpretable estimates, while four exhibited catastrophic structural failures due to logarithmic singularities, unit incompatibilities, and initialization sensitivity. The Bias-Variance Cone framework, representing the first application of asymptotic bias-variance decomposition to fundamental diagram calibration, establishes that free-flow speed parameters are reliably identified within approximately 30 days, whereas density and shape parameters require 100 to 182 days, providing the first quantitative minimum data requirements in the literature. Collectively, the findings provide an empirical identifiability ranking of 17 fundamental diagram models, a validated minimum data sufficiency framework, and a replicable, automated platform for large-scale stochastic fundamental diagram calibration.
Eben Afrifa-Yamoah
Fri 13 Nov
stat.APcs.LGstat.ME
pdf html
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.77
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.72
- interest Causal inference and identification in econometrics 0.80
Abstract
Modelled health estimates such as those of the Global Burden of Disease (GBD) study are increasingly used in secondary analyses designed for observed data. However, trajectory clustering typically ignores published uncertainty, selects the number of clusters using criteria blind to it, and assigns countries to groups with unwarranted certainty. We propose an uncertainty-aware framework that separates trajectory level from shape, quantifies membership probabilities by perturbing estimates within their uncertainty intervals, selects clusters based on stability under uncertainty, and examines the roles of development, region, and measurement quality. Applied to age-standardised cardiovascular disease DALY rates for 201 countries from 1990-2023 (GBD 2023), the framework identified two stable trajectory types, whereas the gap statistic selected three. The additional cluster was poorly supported (median membership probability 0.70) and disappeared once uncertainty was considered. The retained clusters had similar burdens in 1990 but diverged markedly over time: a stalled-improvement group of 78 countries (median change $-8\%$, 40 of them in sub-Saharan Africa) and a sustained-decline group of 92 countries ($-52\%$). Thirty-one countries remained unclassified because their memberships were uncertain. Results were reproduced using mortality rates, a 2000-2023 analysis window, and bootstrap resampling. Among countries with the narrowest uncertainty intervals, a third well-supported trajectory emerged in former Soviet and Eastern European countries. These findings show that uncertainty affects both the number and composition of trajectory clusters. Ignoring it risks over-interpreting unstable groups and overstating confidence in country classifications. Incorporating published uncertainty and measurement quality should become standard practice when clustering modelled health estimates.
Zhengyun Nie, Anant Shyam, Brian Bullins
Wed 18 Nov
math.OCcs.LGstat.ML
pdf html
- interest Computational methods for heterogeneous agent and dynamic economic models 0.79
- favorites near “Monte Carlo Geometry Processing” 0.80
- reading convex optimization, stochastic gradient, sequential, begin
Abstract
We consider the fundamental problem of parallel stochastic convex optimization (SCO), whose aim is to approximately minimize a convex objective when given the ability to query a moment-bounded stochastic gradient oracle multiple times in parallel per round. Whereas most previous results have focused on the variance-bounded Euclidean ($\ell_2$-norm) setting, meaning the objective is Lipschitz with respect to the $\ell_2$-norm over an $\ell_2$-bounded feasible set in $\mathbb{R}^d$, in this work we study parallel SCO in $\ell_p$ settings for $2 \le p < \infty$ under a heavy-tailed noise model. We begin by introducing a new class of $\ell_p$-optimal-distance-convergence (ODC$_p$) algorithms, drawing on prior stochastic bias-reduced techniques [ACJ+21], which combine with multilevel Monte Carlo methods to efficiently estimate the optimum of uniformly convex objectives. As a consequence of our ODC$_p$ algorithms, we first establish rates for projection-efficient $\ell_p$-Lipschitz optimization, composite convex minimization, and efficient $\ell_p$-proximal point estimation. Combining our ODC$_p$ algorithms with a new $\ell_p$-generalized Gaussian-based stochastic gradient estimator ($p$-GREG), we provide two efficient $\ell_p$ parallel algorithms, EpochDA and $\ell_p$-AC-SA, which achieve optimization error $\epsilon > 0$ with at most $\tilde{O}(d^{\frac{p}{2p + 2}} \epsilon^{-\frac{p}{p + 1}})$ query depth, improving upon the sequential heavy-tailed stochastic query depth of $O(\epsilon^{-p})$ for $\epsilon \leq \tilde{O}(d^{-1/(2p)})$. We also establish dimension-dependent lower bounds of $\Omega\bigl(\min\{\epsilon^{-p},d^{p/(3p-2)}\}\bigr)$ adaptive rounds for nonsmooth convex optimization over a subset of the unit $\ell_p$ ball. Our results improve upon the $\Omega(\min\{\epsilon^{-p},\epsilon^{2/3}d^{1/3}\})$ bounds of [DG19], when adapted to the same feasible set.
Weiyan Mu, Shifeng Xiong, Yuming Shen
Mon 16 Nov
stat.MLcs.LGmath.STstat.TH
pdf html
- interest Semiparametric inference and debiased machine learning 0.86
- favorites near “Distributional Random Forests: Heterogeneity Adjustment and Multivariate Distributional Regression” 0.84
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.78
Abstract
The scale parameter in kernel ridge regression plays a pivotal role in modeling the correlation structure of data. In modern applications, sampling density and the underlying correlation range often change simultaneously, making a fixed-scale assumption highly restrictive. However, establishing asymptotic theories under varying scale parameters is fundamentally challenging. To address this gap, this paper develops a unified theoretical framework for kernel ridge regression under varying scale parameters. We establish that, on uniform tensor grids with product exponential kernels, the consistency of the training-point risk holds if and only if the effective local sample size diverges to infinity. This sharp criterion universally covers all regimes along arbitrary deterministic scale sequences, seamlessly unifying the long-range, fixed-scale, and short-range. Furthermore, we provide an explicit, signal-independent construction of a deterministic regularization penalty that guarantees this consistency. Numerical experiments illustrate the criterion and the corresponding penalty across all the three regimes.
My Thi Diem Phan, Tuyen Trung Truong, Hoai Phuong Ha, Dat Thanh Nguyen
Fri 13 Nov
econ.GNcs.LGq-fin.ECstat.AP
pdf html
- reading price, prices, crisis, operators
- favorites near “Eight Centuries of Global Real Interest Rates, R-G, and the 'Suprasecular' Decline, 1311-2018” 0.74
- interest Computational methods for heterogeneous agent and dynamic economic models 0.73
Abstract
After the 2021-2022 energy crisis, day-ahead prices in Nordic bidding zones no longer move together. We ask whether this divergence also changes the information carried by prices. The answer matters for integrating variable renewable energy (VRE), because it can help identify where wind and solar forecasts are worth producing, publishing, and using. We measured the co-movement of each Norwegian and Swedish zone with Germany-Luxembourg (DE-LU) in 2018-2025. We then used a single model, selected on 2024 data, to forecast 2025 prices at 11:00 on the previous day and tested which inputs improved accuracy in each zone. The zones moved alike until 2019, but co-movement diverged in Sweden in 2020 and in Norway in 2021, and the split has persisted. In 2024, four southern zones co-moved strongly with DE-LU (intraday correlations of 0.54-0.71), while four northern zones co-moved weakly (0.25-0.30). Yet price information did not follow this split. Previous-day neighbouring prices reduced mean absolute error by at most 2.2% in any zone, equally in coupled and isolated zones. Removing the transmission system operators' wind and solar forecasts increased mean absolute error by at most 6% in Norway, 8-16% in Sweden, and 31% in DE-LU, and the ordering held in 2024. Within Norway, the value of these forecasts showed no detectable dependence on reservoir filling. The divergence in co-movement was therefore not matched by a divergence in price information, which differed mainly between national generation systems. Thus, a zone's coupling to the continent does not indicate the value of VRE forecasts there.
Mengxi Gao, Gareth O. Roberts, Andi Q. Wang
Mon 16 Nov
stat.COmath.PR
pdf html
- reading processes, inequalities, series, markov processes
- interest Computational methods for heterogeneous agent and dynamic economic models 0.76
- favorites near “Central Limit Theorems for Smooth Optimal Transport Maps” 0.85
Abstract
We study the effect of lifting on continuous-time Markov processes with potentially subgeometric convergence rates. Using a series of weak Poincar\'e-type inequalities, we derive a bound on the maximum acceleration that can be obtained through lifting. We present several examples to explore the sharpness of our derived bound, where the resulting bound has the same square-root order as in the spectral gap setting. Our results provide a framework for understanding the potential and limitations of non-reversible lifting beyond geometric convergence.
Lyuou Zhang, Boxiang Wang, Lulu Wang, Wen Zhou, Hui Zou
Fri 13 Nov
stat.MEstat.ML
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.80
- favorites near “Variational Bayes under Model Misspecification” 0.87
- interest Kernels, operators and functional data 0.79
Abstract
Learning heterogeneity from high-dimensional data has become a fundamental task across many data-driven fields. Cluster analysis is a widely used unsupervised tool for this purpose, and many existing methods have been adapted to high-dimensional settings through sparsity constraints. In contrast to algorithmic approaches, model-based clustering offers clean theoretical guaranties and computational simplicity, but can be restrictive in practice due to its reliance on specific parametric distributions. We propose a novel latent semiparametric mixture model for clustering, in which the observed data follow a Gaussian mixture after latent coordinate-wise monotone transformations. Furthermore, assuming a sparse optimal clustering rule in the latent space, we develop a new method, CESME, tailored for high-dimensional clustering. By allowing unspecified transformations, our framework becomes substantially more flexible and robust than traditional model-based approaches. The clustering error of CESME is shown to approach the optimal clustering error at a rate matching that obtained when the true transformations are known. Extensive simulations and real data analysis confirm that CESME outperforms a broad range of existing high-dimensional clustering methods, including those designed for single-cell RNA sequencing data. The online supplementary file contains technical proofs, additional discussions, and more numerical results.
Mahdi Nouraie, Samuel Muller
Tue 17 Nov
stat.MEmath.STstat.MLstat.TH
pdf html
- reading come, changes, true, forms
- interest Kernels, operators and functional data 0.81
- favorites near “Fit without fear: remarkable mathematical phenomena of deep learning through the prism of interpolation” 0.78
Abstract
Decorrelation has two forms. A left transform keeps coefficients and changes noise; a right transform keeps noise and changes coefficients. Cholesky whitening alone keeps noise variables inactive when true ones come first; decorrelation preserving coefficients and noise loses information.
Ruopei Fu, Ying Yang
Tue 17 Nov
stat.MLcs.LGmath.STstat.TH
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.80
- interest Language models as statistical objects 0.80
- favorites near “Deep Learning for Individual Heterogeneity” 0.78
Abstract
In this work, we study high-dimensional transfer learning with unaligned ordinal responses, allowing different domain-specific regression coefficients, cumulative links and thresholds. We develop the sample-split and full-target algorithms that pool source and target likelihoods and then estimate a target correction. Under suitable conditions, our algorithms achieve minimax-optimal rates for parameter estimation and excess risk. Simulations favor the full-target algorithm over the sample-split algorithm for finite-sample estimation and prediction. An application to a medical dataset further demonstrates the predictive gains of our full-target algorithm over the target-only and pooled baselines when transferring information from the five-category source data to binary hospital mortality.
Matteo Farnè
Tue 17 Nov
math.STmath.OCstat.OTstat.TH
pdf html
- interest Semiparametric inference and debiased machine learning 0.85
- favorites near “Eigendecompositions of Transfer Operators in Reproducing Kernel Hilbert Spaces” 0.84
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.76
Abstract
We study the structural dimension of a second-order object: the minimum latent dimension that remains necessary relative to a declared class of competing dependence. Building on minimum-rank factor analysis and low-rank-plus-sparse decomposition, we use distances to lower-dimensional classes to describe inference near structural ambiguity. General distance and tangent-cone facts yield confidence statements and local margins. For rank-zero competitors with bounded residual degree, we derive exact maximum-weight $b$-matching formulas, support-slack tangent cones and sparse-spike angles; a shared residual support across a field can produce separation absent pointwise. We give positive-semidefinite brackets, a dimension-free Gaussian likelihood expansion, an exact quotient by the tangent of a low-rank background, and a detection floor from the margin. Equal-angle configurations with different branch multiplicities motivate static and spectral detection conjectures. Joint low-rank/residual profiling and the sharp composite information boundaries remain open.
Yen-hsuan Tseng
Tue 17 Nov
stat.MEstat.CO
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.85
- reading log, fit, gives, higher
- interest Bayesian computation and structural estimation 0.78
Abstract
When everyone reports contact counts for known groups, roster-based aggregated relational data (ARD) can determine individual links. In our sparse circular design with 40 nodes and eight groups, the complete counts fix 35.6% of realized edges on average. Zero counts alone exclude 91.1% of nonedges on average, and the full set of counts raises that share to 91.7%. We assess ranking on the unresolved pairs by the area under the receiver operating characteristic curve (AUC). Uniform weighting of compatible graphs yields higher mean AUC than the direct scores from either a circular latent-position fit or a nuclear-norm fit. In a follow-up analysis, using the same fitted probabilities to weight compatible graphs reverses the comparison for the circular fit. With the generating circular form and interaction strength supplied, its mean AUC exceeds uniform by 0.016-0.019 across three paired density settings. The gain is 0.007 under a stochastic blockmodel, where the circular form is misspecified. Nuclear-norm plug-in scores remain below uniform in mean AUC. For independent Bernoulli edges with fixed probabilities strictly between zero and one, we characterize when an oracle using the true probabilities gives the same edge marginals as uniform weighting for every feasible count array. This occurs exactly when the conditional graph law is uniform on every such array, or equivalently when log odds are additive in node-by-group effects. We recommend reporting fixed edge states, ranking on unresolved pairs, and generating-probability error separately.
Eduardo Pinto, Panagiota Birmpa
Mon 16 Nov
stat.MLcs.LG
pdf html
- reading empirical, little, practice data, viewed
- your work near “Valid Inference with Imperfect Synthetic Data” 0.78
- interest Online learning and statistical learning theory 0.82
Abstract
We study the asymptotic behavior of optimal discriminators associated to $L-$Lipschitz regularized KL-divergences of two probability distributions $Q$ and $P$ that interpolate between the rescaled Wasserstein$-1$ metric by some parameter $L>0$ and the KL divergence. The parameter $L$ plays the role of a Lipschitz constraint and is viewed as a regularization parameter: small values of $L$ enforce a more geometry-aware discriminator, while as $L$ increases the geometric regularization is progressively relaxed. It is a known result [3] that such divergences converge to the KL divergence as $L\to\infty$. In this paper, we characterize the asymptotic behavior of the optimal discriminators as $L\to\infty$ in different regimes: Firstly, under $Q\ll P$ and the Radon-Nikodym derivative $\frac{dQ}{dP}$ is bounded from above and bounded away from zero. Secondly, when $Q$ and $P$ are discrete distributions satisfying $Q\ll P$, and finally when $Q$ and $P$ are non absolutely continuous discrete distributions. This result motivates training strategies which involve discriminators, such as adversarial generative models. In practice, the data distribution is often represented by a discrete empirical measure, while, especially at the early stages of the training, the model distribution may have little or no overlap with the empirical data distribution.
Konstantin Yakovlev
Fri 13 Nov
math.STcs.LGstat.TH
pdf html
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.79
- reading dimensionality, functions, formal, success
- interest Language models as statistical objects 0.79
Abstract
Flow and diffusion-based generative models achieve state-of-the-art quality, but require multiple forward passes at inference time, making them computationally costly. Consistency models overcome this by learning a flow map directly, enabling few-step generation and dramatically reducing the computation needed at test time. However, formal statistical guarantees for such one-step generators are scarce. Existing results either suffer from the curse of dimensionality or rely on impractical loss functions. We investigate whether one-step flow-based models can escape the curse of dimensionality under the assumption that the data distribution is generated by a low-dimensional surface perturbed with Gaussian noise. In this setup, we derive approximation and generalization error bounds for the flow map estimate that circumvent the curse of dimensionality. Our findings provide the first principled explanation for practical success of one-step flow-based generative models.
Mathieu Blondel, Michaël E. Sander, Germain Vivier-Ardisson, Tianlin Liu, Vincent Roulet
Fri 13 Nov
cs.LGstat.ML
pdf html
- favorites near “Sequential Monte Carlo Steering of Large Language Models using Probabilistic Programs” 0.80
- reading functions, moment, attempts, language
- interest Language models as statistical objects 0.79
Abstract
During the supervised learning phase, autoregressive language models are typically trained by minimizing the negative log-likelihood of the training data, which decomposes as a sum of token-level cross-entropies via teacher forcing. This objective is simple and efficient but potentially suffers from so-called exposure bias: a mismatch between the way the model is trained and the way it is used at inference time. Can we instead design sequence-level loss functions that incorporate inference-time decoding procedures, such as greedy decoding or autoregressive sampling? We show that two natural attempts lead to loss functions that are fundamentally flawed: unbounded below. We further show that these two flawed loss functions are intimately connected: using autoregressive Gumbel tricks, the sampling based loss is the expected greedy loss under Gumbel perturbations. Since the pathology stems from the dependence of the decoded sequence on the model parameters, we analyze the update direction obtained by applying a stop gradient on the sampling procedure, leading to a moment matching direction. While it is not the gradient of any loss function, we prove that it defines a valid descent direction in function space for the autoregressive negative log-likelihood. Despite these theoretical insights, the moment matching direction suffers from high variance. We conclude that teacher forcing may well be ``all you need'' for supervised learning of autoregressive models.
Félix Côté, Marie-Pier Côté
Fri 13 Nov
stat.MLcs.LG
pdf html
- reading structural, science, functions, considerably
- interest Online learning and statistical learning theory 0.83
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.77
Abstract
The short duration of general insurance contracts leads to rapid distributional shifts in the data, warranting regular model updates. This process may yield a post-update model that differs considerably from the pre-update model in both predictions and structure. Structural stability is upstream of predictive stability, yet existing structural stability metrics are mostly restricted to simple decision trees. Thus, we first introduce the notion of functional transparency, characterising models with a prediction function that can be expressed as a sum of univariate functions for main effects and bivariate functions for pairwise interactions. Then, we propose the Mean Squared Shape function distance, a metric to quantify the structural stability of functionally transparent models, which are particularly relevant in actuarial science. Using three public auto insurance datasets, we present a benchmark study to demonstrate the usefulness of our metric in identifying a stable and accurate transparent model. Generalised linear models with centred numerical covariates are stable, and explainable boosting machines (EBMs) offer a compromise between stability and performance. We also find that stabilisation strategies, such as sequential modelling and fixed hyperparameters, can substantially improve stability for both EBM and XGBoost models, with negligible impact on performance.
Yuxin Han, Lanjun Cheng, Fengshi Jing
Tue 17 Nov
stat.MEmath.STstat.TH
pdf html
- interest Adaptive data collection and anytime-valid inference 0.85
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.80
- favorites near “Hierarchies of Relaxations for Online Prediction Problems with Evolving Constraints” 0.78
Abstract
An experiment-level failure replaces the joint observation law while preserving the target. We study honest confidence sets when a declared upper bound on failure probability equals the allowed noncoverage probability. For a Uniform support endpoint and a Gaussian mean on fixed compact parameter intervals, every honest deterministic procedure has full-range worst-case expected diameter under arbitrary or absolutely continuous joint failures. Independent randomisation gives sharp minimax risks of $(2+o(1))\log N/N$ for the endpoint in $[1,2]$ and $(2\sigma+o(1))(\log N/N)^{1/2}$ for the mean with known variance $\sigma^2$. A failure distribution shared by many targets forces average clean-state noncoverage to decrease with expected diameter. One-sided local Kullback--Leibler bounds then yield lower bounds with logarithmic factors. The endpoint experiment also has a sharp deterministic frontier as the remaining error budget vanishes. Under exchangeable failures without ties, observation order recovers both randomised benchmark rates deterministically. In a reference-calibrated roster, we derive the randomised minimax reliability rate. A prespecified reference reserve of logarithmic size permits deterministic attainment of the same order on a subclass whose conditionally independent reserved responses have probabilities bounded away from zero and one in every state.
Zimeng Ren, Kun Huang, Yongming Qu, Wanzhu Tu, Xiaoqing Huang
Tue 17 Nov
stat.AP
pdf html
- reading heterogeneity, observational, did, empirical
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.82
- favorites near “Bayesian Semiparametric Model for Sequential Treatment Decisions with Informative Timing” 0.79
Abstract
Subgroup heterogeneity in carryover effects is not routinely evaluated in 2x2 crossover studies, and its inferential consequences remain insufficiently characterized. When the scientific question concerns subgroup-specific carryover, omission of the corresponding interaction causes the fitted model to target a collapsed population-average contrast rather than the subgroup difference of interest. We used simulation to evaluate the finite-sample bias, confidence-interval coverage, Type I error, and power of carryover-by-subgroup interaction inference under an independence-based analysis and REML-based MMRM analyses with CS and UN covariance specifications. Reduced mean models were included as population-average benchmarks. We also applied the models to observational Alzheimer disease treatment-transition sequences from NACC, where carryover was defined operationally as a model-based association with the preceding regimen rather than an identified pharmacologic persistence effect. Full models directly estimated subgroup-specific carryover heterogeneity, whereas reduced models estimated a collapsed carryover contrast that did not characterize this heterogeneity. Under the full mean model, point-estimate bias was small, whereas covariance misspecification affected empirical confidence-interval coverage and Type I error. Reliable inference on subgroup-specific carryover effects requires a mean model that explicitly represents the heterogeneity of interest and a covariance structure appropriate for the repeated-measures data. Failure to address either component can obscure subgroup differences or compromise statistical inference. These results provide practical guidance for evaluating heterogeneous carryover effects in crossover and treatment-transition studies.
Roi Naveiro, William N. Caballero, Juan Maroñas
Tue 17 Nov
stat.MLcs.LG
pdf html
- reading inference, moments, distribution suitable, specifying
- your work near “Valid Inference with Imperfect Synthetic Data” 0.78
- interest Adaptive data collection and anytime-valid inference 0.81
Abstract
Adversarial machine learning has revealed that statistical models can be manipulated through poisoned data, and Bayesian inference is no exception. Earlier work has shown that subtle poisoning attacks can drive the posterior toward a distribution of the attacker's choosing. Such posterior attraction attacks require a fully specified target posterior; however, attackers typically care about only a few posterior quantities or, more consequentially, about a downstream decision. Herein, we study how an attacker can alter Bayesian inference and decisions by deleting or replicating a few observations, chosen to minimize the maximum mean discrepancy between the resulting posterior and a target distribution. A suitable kernel lets the attacker steer selected posterior moments without specifying the rest of the target. Because Bayes actions depend on the posterior only through expected utilities, which are themselves posterior moments, the same approach can target a decision directly. We construct the attacks with a gradient-based heuristic that requires only posterior samples, and analyze its convergence. Experiments with regression and hierarchical models compare moment-based and full-posterior attacks, and illustrate how deleting a handful of observations can flip a Bayesian decision.
Michael Wieck-Sosa, Cosma Rohilla Shalizi
Tue 17 Nov
stat.MLcs.LGstat.COstat.ME
pdf html
- reading conditional density, massive, good, dimensionality
- favorites near “PAC-Bayes Compression Bounds So Tight That They Can Explain Generalization” 0.81
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.76
Abstract
Neural posterior estimation (NPE) is a popular Bayesian simulation-based inference (SBI) approach that uses neural networks trained on simulated data--parameter pairs to approximate the posterior given a user-chosen prior. For high-dimensional observations, NPE methods typically approximate the conditional density of the parameter given a user-chosen or learned summary statistic, rather than the entire dataset, due to the curse of dimensionality. However, both approaches face challenges: hand-crafting good summaries for complex models requires considerable effort from scientists, while learning them from massive datasets can be prohibitively computationally expensive. To address these challenges, we propose approximating the conditional density of a parameter given low-dimensional sketches of the data. We show that, for a model with a $p$-dimensional parameter, only $2p+1$ random sketches are needed for posterior consistency. We compare with state-of-the-art embedding networks for NPE, and we find that our method \emph{matches} the best embedding network's performance while \emph{increasing computational efficiency}.
Lu Mao
Fri 13 Nov
stat.MEstat.ML
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.80
- favorites near “Bayesian Semiparametric Model for Sequential Treatment Decisions with Informative Timing” 0.83
- reading history, semiparametric, doubly robust, doubly
Abstract
We consider individual prediction of general summaries of a multistate trajectory, such as time alive out of hospital or time spent in remission. Right censoring complicates calibration because these outcomes may be only partially observed. We develop a conformal prediction approach for scalar and vector summaries over a fixed clinical horizon supported by follow-up. For each candidate region, we define the coverage determining time as the earliest time at which the observed history establishes whether the outcome lies in the region, even if the outcome itself is still unknown. We weight observed coverage decisions by the inverse probability of remaining uncensored to that time and augment the estimator with conditional coverage probabilities for unresolved histories. For a fixed region, the coverage estimator is doubly robust with respect to the censoring and conditional coverage models and attains the semiparametric efficiency bound when both are consistent at suitable rates. For the calibrated region, we derive a finite-sample lower bound on the coverage probability and establish asymptotic validity if either model is correct. In simulations, using partial histories yields narrower prediction intervals at comparable coverage. We illustrate joint prediction of days in and out of hospital in the HF-ACTION trial and prediction of relapse-free time in the German Breast Cancer Study Group cohort.
Subir Hait
Tue 17 Nov
stat.MEecon.EM
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.80
- reading adjustment, rd, feedback, causal
- interest Reinforcement learning, policy learning and off-policy evaluation 0.76
Abstract
Hospitals use repeated risk scores to decide when to send an alert. Some evaluations summarize these scores by their maximum and use the maximum as the running variable in a regression discontinuity (RD) design. The maximum shows correctly whether a patient was ever alerted. However, it can include scores recorded after an earlier alert has already changed care. In this paper, we study what the maximum-score RD estimates when an earlier intervention changes a later score. We use a two-assessment model and derive a formula that splits the RD contrast into the causal effect and a composition term. When feedback moves scores in one direction, the composition term equals the relative jump in the density of the maximum at the cutoff times a prognostic contrast. Therefore, an exactly continuous density means no bias, and the population density jump bounds the bias. Valid inference on the jump turns this into a confidence bound under a stated sensitivity assumption. Within this model, bias behind a continuous density requires feedback in both directions. For this case, we bound bias by total boundary movement and construct conservative confidence sets for binary outcomes under explicit sensitivity and smoothness assumptions. We give closed forms for independent uniform scores and for correlated Gaussian scores. In one example, a true risk reduction of four percentage points appears as a one-point increase. Simulations confirm the formulas. Narrower bandwidths, robust bias correction, and machine-learning covariate adjustment do not remove the bias. A fuzzy RD on the first score recovers the causal effect when early and late treatment have the same effect. Our results concern how the running variable is built. We do not assess any published clinical study.
Arya Mazumdar, Prateeti Mukherjee, Harsh Vardhan
Thu 19 Nov
stat.MLcs.DScs.LGmath.STstat.TH
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.79
- interest Online learning and statistical learning theory 0.83
- favorites near “Orthogonal Statistical Learning” 0.86
Abstract
We revisit the classical problem of learning general Gaussian halfspaces in the presence of label noise, with particular focus on the large-bias regime, where information-computation gaps are conjectured in several standard label-noise models. Concretely, we present and analyze the rare-tail mean estimator, an extremely simple one-shot statistic that averages only the observed minority class. In the large-bias regime, classical Averaging (Servedio, 1999) and related global first-moment statistics are provably suboptimal, while the rare-tail mean estimator recovers the target halfspace direction at the information-theoretic sample rate (modulo log factors) under a broad class of label noise models, including probit noise and margin-dependent noise with very slow decay away from the boundary. Notably, our guarantees allow labels arbitrarily close to the decision boundary to be nearly random, while still giving strong directional recovery from a small number of samples. The estimator also cleanly interpolates between the noiseless and noisy regimes: without noise, it attains the same sample rate as classical separability-based methods for realizable halfspaces, and under low noise, or noise that decays away from the boundary, it keeps this rate even after linear separability has been destroyed. As an application, we consider a two-stage curriculum in which source samples, whose noise may be arbitrarily close to $1/2$ near the boundary and decay very slowly away from it, precede target samples, whose noise flips every label at the same rate, i.e., random classification noise (RCN). Since both stages share the same Bayes optimal halfspace, our estimator supplies the directional warm start needed by existing refinement routines, and together they learn the RCN target at the information-theoretic sample rate in the large-bias regime, up to log factors.
Kisung You
Fri 13 Nov
stat.ME
pdf html
- interest Kernels, operators and functional data 0.84
- favorites near “Eigendecompositions of Transfer Operators in Reproducing Kernel Hilbert Spaces” 0.82
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.76
Abstract
Directional distributions can have the same normalized mean direction while differing substantially in shape. Comparing them requires a distance whose response to geometric displacement and probability overlap matches the statistical question. We examine ten bounded metrics or metric families for probability measures on the sphere, scaling each to have maximum value one, covering transport, sliced transport, kernel and energy distances, and comparisons based on probability mass. We explain which metrics describe weak convergence, which admit Hilbert-space norm representations, and which pairs attain the maximum distance. Examples and counterexamples show how these distinctions affect empirical comparisons and numerical approximations. A summary table and relationship diagram support the choice of a distance. The discussion separates exact metrics from finite representations, unbiased estimators, and regularized transport costs.
Sikun Xu, Raphael Thomadsen, Dennis J. Zhang, Heng Zhang
Wed 18 Nov
cs.LGcs.CLstat.MEstat.ML
pdf html
- favorites near “Data Science at the Singularity” 0.83
- your work near “Valid Inference with Imperfect Synthetic Data” 0.79
- interest Semiparametric inference and debiased machine learning 0.82
Abstract
Public LLM leaderboards guide model evaluation and adoption, yet the reported performance of top-ranked models can be systematically overstated. This leaderboard winner's curse arises because ranking noisy benchmark scores rewards sampling luck alongside genuine ability. The same selection effect can distort scores throughout the leaderboard, a phenomenon we call the rank-induced bias. We develop an $m$-out-of-$N$ bootstrap to correct for this bias. It assumes a fixed model pool with independent, identically distributed bounded tasks while allowing arbitrary within-task correlation among models. Under exact population tiers separated by positive gaps, we prove that the bootstrap reduces $\Theta(N^{-1/2})$ ranking bias to expected residual bias $o(N^{-1/2})$. Across 14 benchmarks from 5 benchmark families, the rank-1 adjustments range from approximately 0.09 to 3.86 percentage points. In our SWE-bench Verified snapshot, the tied leaders receive an adjustment of 0.90 percentage points. Such overstatement is large enough to make the apparent winner statistically fragile. Thus, we use the same bootstrap to further quantify the intervals of plausible rank positions for each model and group models into tiers. We establish sufficient conditions for consistent recovery of population tiers. On SWE-bench Verified, this procedure places 13 of 173 models in the leading tier. We combine the score adjustment and rank intervals in the Rank Uncertainty and Leaderboard Evaluation Report (RULER). We validate its finite-sample performance using sample splitting and calibrated simulations, and advocate publishing RULER alongside raw rankings so readers can assess both reported performance and the precision of the ordering. RULER also provides guidance for benchmark design that accounts for selection effects and ranking uncertainty.
Yilin Li, Jiangshan Zhang, Jiming Jiang
Wed 18 Nov
stat.ME
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.82
- interest Causal inference and identification in econometrics 0.81
- reading causal, doubly robust, empirical, inference
Abstract
We develop a framework for small area causal inference (SACI). Unlike the traditional doubly robust (DR) estimation approach, the new framework is based on the empirical best prediction (EBP) method. EBP takes into consideration not only the bias but also the variance factor, which is often a more dominant component in the mean squared prediction error (MSPE). Under a linear mixed model for SACI, we establish consistency and asymptotic normality of the maximum likelihood estimator of the model parameters. More relevant to our main objective, we establish asymptotic optimality of the EBP in terms of minimizing the MSPE of the area-specific causal-effect (ASCE) predictor. Moreover, in view of the traditional interest in causal inference, we establish asymptotic normality of the EBP of the ASCE , which is used to construct prediction intervals for the ASCEs. We carry out extensive simulation studies to investigate finite-sample performance of the proposed EBP and its comparison with the traditional DR and other existing methods. Our simulation results show that the proposed EBP method has significant advantage over the comparing methods in terms of MSPE and coverage probability and length of the resulting prediction interval. A real-data example from agricultural-animal genetics is discussed.
Sebastian Fajardo, Philip Riris, María Vélez, Natalia Álvarez Correa, A. Jacobo Hidalgo, Luis Felipe Navarro, et al.
Mon 16 Nov
stat.AP
pdf html
- favorites near “When Did Growth Begin? New Estimates of Productivity Growth in England from 1250 to 1870” 0.78
- reading technology, growth, climatic, technological
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.71
Abstract
Human communities produce technology in almost every environment, but the conditions that enable technologies in the tropics are not well understood. We test two competing hypotheses---that technological emergence is triggered by environmental stress and that technology tracks growth in effective populations---by integrating Holocene archaeological and paleoclimate evidence. We use archaeological derived radiocarbon dates from Colombia as a proxy for human activity, including dating plant management, pottery, metallurgy, and raised fields. For palaeoclimate, we use two El Ni\~no--Southern Oscillation (ENSO) reconstructions: Laguna Pallcacocha and Laguna El Junco in Ecuador. A Bayesian changepoint at 3575 cal BP (95% highest posterior density interval (HPDI) = 3811--3329) marks a ~13-fold increase in the growth rate of pre-Hispanic human activity ($r_2$/$r_1$ median = 13.3, 95% HPDI = 8.1--22.2). Plant management and pottery predate this shift by more than a millennium, whereas metallurgy and raised fields cluster within the early acceleration phase, during relative climatic and demographic stability. A Bayesian log-linear regression of human activity on ENSO frequency and intensity after the demographic changepoint reveals an asymmetric covariation: reduced ENSO frequency and intensity suppress activity by 58% and 21%, whereas ENSO frequency raises activity modestly (~11--13%). Therefore, ENSO regulates demographic activity rather than triggering technological emergence. These patterns suggest that pre-Columbian communities may have innovated when they could compute---when an effective population sat above a critical threshold and stability windows permitted combinatorial search---rather than when they were forced to by environmental stress, expressed as demographic or climatic variability.
Zhenisbek Assylbekov
Wed 18 Nov
cs.LGstat.ML
pdf html
- reading counterexamples, higher, make, recipe
- interest Kernels, operators and functional data 0.78
- favorites near “Mean field games and applications” 0.80
Abstract
e-means (Bessa et al., 2026) combines exponential averaging and cluster revival. We give two counterexamples to its published recipe with higher final loss than standard k-means (Lloyd, 1982) from identical initial centers. Mass-based revival can destroy a global optimum and make the ratio of expected losses unbounded even with k-means++ seeding.
Athanasios Christou Micheas
Fri 13 Nov
math.PRmath.STstat.TH
pdf html
- reading theory, value function, functional, processes
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.76
- interest Bayesian computation and structural estimation 0.79
Abstract
Using point process theory we propose and investigate a probabilistic identification of zeros of a deterministic real valued function. We construct a random atomic driving measure for a Cox point process whose mass at a candidate location depends explicitly on the value of the function at that location. We study the properties of the constructed random measure, including its exact asymptotic properties, its Laplace functional and related random processes. We also give an important application involving zero detection for Riemann's {\xi} function.
Hana Sagor, Edward L. Boone, Ryad A. Ghanam
Wed 18 Nov
stat.AP
pdf html
- favorites near “Eight Centuries of Global Real Interest Rates, R-G, and the 'Suprasecular' Decline, 1311-2018” 0.78
- interest Computational methods for heterogeneous agent and dynamic economic models 0.76
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.75
Abstract
This paper investigates whether neural networks can infer fractional option prices and the Hurst parameter directly from sequences of fractional Gaussian returns, without separately estimating the latent parameters for each new sequence. Feedforward neural networks (FNNs) and long short-term memory (LSTM) networks are trained to jointly predict European call prices, put prices, and the Hurst parameter from 252-return sequences. A controlled simulation varies the total sample size from 5,000 to 500,000 sequences and evaluates point prediction, empirical residual-based prediction intervals, and computational performance. The neural-network models are also compared on a common simulated test set with Bayesian joint estimation and exact Gaussian maximum likelihood estimation. In that four-method comparison, the LSTM achieves the lowest Hurst-parameter root mean squared error (0.033) and option-price root mean squared error (1.305). After training, neural-network predictions are substantially faster than repeated parameter estimation, although the methods quantify uncertainty in different ways. Applications to five financial return series illustrate differences between data-driven and model-based fractional price estimates. The results support sequence-based learning as a computationally efficient complement to statistical inference under the assumed fractional model; they do not establish out-of-distribution accuracy or accuracy against observed traded-option prices.
Ishaq Aden-Ali, Yeshwanth Cherapanamjeri, Mikael Møller Høgsgaard, Kasper Green Larsen, Nikita Zhivotovskiy
Tue 17 Nov
math.STcs.LGstat.TH
pdf html
- favorites near “Orthogonal Statistical Learning” 0.87
- interest Kernels, operators and functional data 0.81
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.73
Abstract
We study the problem of estimating the mean of a random vector in $\mathbb R^d$ under $\ell_p$ norms, assuming only the existence of a covariance matrix. In the Euclidean norm, the sample mean is already optimal in expectation, and robust estimators are usually needed only to obtain high-probability tails. Our results show that this picture changes for $\ell_p$ norms with $p>2$: robustification is already necessary even for the expected risk. We prove that the minimax risk exhibits a sharp phase transition, controlled by the interaction between the sample size $n$, the dimension $d$, and the norm parameter $p$. In the large-sample regime, the classical Gaussian rate remains achievable. In the complementary regime, heavy-tailed distributions create a genuinely non-Gaussian obstruction. Furthermore we show that the simple estimator: coordinate-wise median of means whose number of buckets chosen according to the geometry of the loss, optimally tracks this phase transition in its expected risk minimax. We also show that natural projection-based median-of-means estimators for general norms, despite achieving the optimal confidence trade-off relative to the sample-mean benchmark of Lugosi and Mendelson (Probab. Theory Relat. Fields, 2019), can be polynomially suboptimal for $\ell_p$ norms even in the constant-probability regime.
Boumediene Hamzi, Houman Owhadi
Fri 13 Nov
stat.MLcs.CEcs.LGmath.DSnlin.PS
pdf html
- favorites near “Eigendecompositions of Transfer Operators in Reproducing Kernel Hilbert Spaces” 0.86
- interest Kernels, operators and functional data 0.82
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.74
Abstract
When do mode energies preserve regime information as measurement quality changes? Stochastic Kernel Mode Decomposition (sKMD) fits joint regime laws on signed KMD mode coordinates and transfers the reference mode posterior to regime probabilities and conditional reconstructions. For centered Gaussian regimes with independently varying known noise variances, fixed coordinate energies retain the full regime posterior precisely when all classes share their cross-covariances and class variances can differ only at isolated vertices of the covariance graph. We extend the characterization to nonzero means and show that freely varying correlated noise forces a fixed linear log-odds decoder to retain the full quadratic feature span even for two distinct centered regimes. At fixed quality, a three-regime example has no sufficient two-energy basis. Its smallest Bayes log-loss gap over bases of condition number at most $\kappa$ has sharp order $\kappa^{-1}$. Exact Bayes energy posteriors supply numerical gap estimates for specified bases. Additional bounds relate feature compression, independently learned mode spans and sensor-noise error to log loss. Two controlled experiments with 64 independent panels each show that signed information can improve prediction over energy summaries, while integration gives small gains over strong signed controls and depends on how the competing point rule is calibrated and how accurately observation noise is measured.
Bogdan Butyrin, Eric Moulines, Alexey Naumov, Sergey Samsonov
Fri 13 Nov
math.PRmath.STstat.MLstat.TH
pdf html
- interest Computational methods for heterogeneous agent and dynamic economic models 0.77
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.77
- reading linear stochastic, bound holds, bootstrap procedure, bound error
Abstract
We study how quickly the normalized Polyak--Ruppert average approaches its Gaussian limit in linear stochastic approximation with bounded i.i.d.\ coefficients. For step sizes decaying as $k^{-\gamma}$, with $1/2\le\gamma<1$, we prove an upper bound of order $n^{\gamma-1}$ on the approximation error over convex sets in fixed dimension. A matching lower bound holds for every fixed model. The fastest rate in this range is $n^{-1/2}$, attained at $\gamma=1/2$. We also obtain an $O(n^{-1/2})$ bound on the error of the recursive multiplier bootstrap procedure.
Andrea Da Col, Cristian R. Rojas
Tue 17 Nov
eess.SPecon.EM
pdf html
- reading state, root, children, actually
- interest Kernels, operators and functional data 0.80
- favorites near “The Wealthy Hand-to-Mouth” 0.81
Abstract
We study recursive estimation of a constant state vector by a hierarchical network of asymmetrically informed linear estimators, modeled as a directed tree. Only the leaves observe the regression signal, whereas every other estimator learns by word of mouth from noisy signal estimates relayed by its children. Authority within the hierarchy is expressed through common prior information: after each round, every estimator adopts the estimate and error covariance broadcast by the root. The more accurate this common information, the larger the effective noise variance on local observations, which makes estimation intrinsically self-defeating. The resulting performance is the subject of this paper. We show that recursive least-squares (RLS) estimators do learn the state in mean square, albeit slowly. Under strong persistency of excitation, the broadcast error covariance decays polynomially in time, at a rate that we characterize by upper and lower bounds of matching order. The exponent is determined by the smallest number of relays between a leaf and the root, and it decreases exponentially with this number.
Anasse Essalih, Reda Khoufache, Laurent Durville, Mohamed-Cherif Rahal, Mustapha Lebbah, Hanene Azzag, et al.
Tue 17 Nov
stat.MLcs.AIcs.LG
pdf html
- favorites near “Variational Bayes under Model Misspecification” 0.87
- reading variational, competitive results, co-clustering, clustering performance
- interest Language models as statistical objects 0.77
Abstract
We introduce LBM-VAE, a novel framework that integrates Latent Block Models (LBMs) into Variational Autoencoders (VAEs) to jointly cluster observations and latent dimensions. Beyond its clustering capabilities, LBM-VAE provides a block-structured latent representation that naturally supports disentangled representation learning. In contrast to conventional approaches that associate each factor of variation with an individual latent dimension, our framework allows factors of variation to be encoded across groups of latent dimensions. Experimental results demonstrate the promising clustering performance of LBM-VAE, while achieving competitive results in disentangled representation learning. LBM-VAE highlights the potential of combining co-clustering and variational inference to learn structured and interpretable latent representations.
Nikola Banić, Neven Elezović
Fri 13 Nov
math.STmath.PRstat.TH
pdf html
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.77
- interest Macroeconometrics and time series 0.81
- favorites near “Optimal uniform convergence rates and asymptotic normality for series estimators under weak dependence and weak conditions” 0.84
Abstract
We study the cyclic discrete total variation obtained by adding the wrap-around edge to the usual open-chain statistic. Under the uniform multinomial law the cyclic symmetry recombines the same per-edge, adjacent, and disjoint covariance blocks used for the linear statistic, but with different multiplicities, and this simplifies the variance: the bin-independent leading coefficient \(2A+4B\) carries over, the constant-order correction collapses \emph{exactly} to \(\beta_c(n)=\beta_1 n\) (no \(n^0\) or \(1/n\) tail; obtained directly here, and already reached in Part~2 through the transfer identity), and a conditional second-order coefficient \(\gamma_c(n)\) is given. The exact finite-\(N\) cyclic--linear transfer identity of Part~2, read order by order, both delivers these and, as its main consequence for the \emph{linear} statistic, explains the traveling mode of the linear variance: comparison with the (first-order) mode-free symmetric cyclic profile shows that deleting one edge introduces a rising contribution whose balance against the falling cyclic bulk creates a ridge at \(n\asymp\sqrt{K_0N}\) with local optimizer \(K_0=C_1/\beta_1\). Uniform control of three ridge blocks proves this optimizer on every compact ridge window, upgrading the previously heuristic \(\sqrt N\) mode law to a theorem; the global localization of the exact maximizer remains open. We also record the corresponding cyclic reference-centered statistic under a fixed alternative and the leading Nabeya variance formula for the matched non-uniform cyclic statistic. Throughout, ``exact'' refers to the asymptotic coefficient in \(N\); the operative symmetry is cyclic stationarity (rotation invariance), not exchangeability.
Daniel Ageev, Yariv Aizenbud
Thu 19 Nov
math.STstat.TH
pdf html
- reading laplacian, eigenvectors, operator, eigenvalues
- favorites near “Monte Carlo Geometry Processing” 0.80
- interest Kernels, operators and functional data 0.74
Abstract
The first eigenvalues and eigenvectors of the unnormalized graph Laplacian of clean samples from a manifold, suitably scaled, approximate those of a weighted Laplace-Beltrami operator. However, the samples are usually observed with noise and therefore need not lie on the manifold. We build the clean and noisy graph Laplacians, with bandwidth $\varepsilon$, on the clean samples and on the noisy observations, where the noise is arbitrary but bounded by $\delta$. We show that, with high probability, the first eigenvalues and eigenvectors of the noisy graph Laplacian are close to those of the clean one: the eigenvalues within $O(\delta/\varepsilon)$, and the subspaces spanned by the first eigenvectors within $O(\sqrt{\delta/\varepsilon})$, whether or not the eigenvalues are simple. Consequently, if $\delta$ is small enough relative to $\varepsilon$, the noisy graph Laplacian approximates the weighted Laplace-Beltrami operator at the same rate as the clean one. We also show that, for large noise, we can get the same rate as the clean graph Laplacian by first using a denoiser, and then building the graph Laplacian on the denoised observations.
Jinran Wu, Geoffrey J. McLachlan
Wed 18 Nov
stat.COstat.AP
pdf html
- interest Language models as statistical objects 0.80
- reading standard errors, kernels, implications, deep gaussian
- your work near “Timing as an Action: Learning When to Observe and Act” 0.76
Abstract
Deep mixture models extend finite mixture models through hierarchical latent factor structures with shared mixture parameters. This focused review examines deep Gaussian mixture models (DGMMs), robust deep mixture models (RDMMs), deep skew-$t$ mixture models (DStMMs), and mixed deep Gaussian mixture models (MDGMMs) within a common statistical framework. We compare their pathway distributions, measurement models, and latent architectures, highlighting the implications for likelihood evaluation and inference. For continuous measurements, we discuss Gaussian conditional calculations and the scale-mixture methods underlying Student-$t$ and skew-$t$ pathways. For mixed measurements, we consider response-specific likelihoods and numerical integration for discrete variables. Alongside the review, we introduce \texttt{MixedDeepMM}, an R package providing a unified interface for fitting and evaluating deep mixture models with Gaussian, Student-$t$, and skew-$t$ kernels, supporting continuous and mixed measurements and single- and two-head architectures. We illustrate its use on the UCI Wine and Cryotherapy datasets, reporting clustering measures with standard errors across repeated fits and visualising consensus partitions. For mixed data, Gower silhouette widths complement external clustering agreement measures. We conclude by discussing computational challenges, model selection, and uncertainty assessment.
Tobias B. Konitzer, David M. Rothschild
Thu 19 Nov
econ.EM
pdf html
- reading demographics, match, advertising, imply
- your work near “Valid Inference with Imperfect Synthetic Data” 0.79
- interest Causal inference and identification in econometrics 0.76
Abstract
Third-party identity-resolution vendors commonly emphasize match rate, the share of submitted records they can link to identifiers for advertising activation, as their core metric of quality. Yet a high match rate does not necessarily mean that those matched identifiers belong to the submitted records (i.e., the intended targets). In this paper we explore match quality as an alternative metric to match rate, utilizing a unique data set to evaluate match quality. We are able to compare two demographics, age and gender, from a submitted record (voter files) to the same demographics in self-reported surveys matched to mobile advertising identifiers linked together by two leading identity-resolution vendors. The data and matching was from 2019-20. There is substantial ambiguity in match quality of the returned records. First, both vendors' activation files contain multiple device identifiers per match. Second, match quality among audited pairs (Vendor A) and for explicitly labeled individual-level matches (Vendor B) is only 21.8% and 23.6%, respectively. Utilizing a mixture model that assumes that correct links always agree on these two demographics and incorrect links agree 10% of the time, these rates imply conditional identity-level precisions of approximately 15% and 13%, respectively. This paper utilizes identity-resolution products that are several years old, but the question we address remains fundamental: match rate alone cannot establish whether advertisers are reaching the people they intended. Evaluating identity resolution requires independent evidence of match quality alongside match rate.
Zhengyuan Fan, Zhonghua Wu, Tianyu Cheng, Qun Chen
Thu 19 Nov
stat.MLcs.AI
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.77
- favorites near “Dynamically Optimal Treatment Allocation” 0.80
- interest Reinforcement learning, policy learning and off-policy evaluation 0.79
Abstract
Rubric-based reinforcement learning constructs rewards for aligning large language models with human preferences, typically by equally aggregating continuous scores that a judge assigns to individual criteria. However, scoring heterogeneity across criteria can distort candidate rankings. Existing work has focused on improving judges, learning criterion weights, or refining aggregation, whereas the interaction between scoring heterogeneity and ranking remains insufficiently understood. We formalize this heterogeneity as \emph{reward gain imbalance}: each criterion enters the aggregated reward with its own reward gain, the gap between its expected scores on satisfied and unsatisfied responses. Counterintuitively, our experiments show that removing this imbalance, even with ground-truth labels, lowers ranking accuracy. To explain this, we derive an exact decomposition of the reward difference between any two candidates into a base satisfaction signal, a reward gain imbalance term, and a residual term, showing that reward gain imbalance can attenuate or amplify the base signal, and finding that misrankings usually concentrate on pairs where it attenuates. Guided by this analysis, we propose the adaptive \textbf{D}ual-\textbf{C}enter \textbf{C}lipping aggregation (DCC), which bounds each criterion's contribution to the reward difference instead of simply removing the imbalance. Using only the scores of the current candidate group, it determines a clipping interval that imposes a common upper bound on per-criterion score differences between candidates, limiting the excessive influence of extreme scores while preserving continuous judgment information....
Molei Liu, Tian Jiang, Keyao Zhan, Zirui Zhang, Xuan Wang
Tue 17 Nov
stat.ME
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.80
- favorites near “Deep Learning for Individual Heterogeneity” 0.78
- reading counterfactual, treatment, partially identified, good
Abstract
Surrogate endpoints can accelerate treatment evaluation, but their validity is challenging to establish because the joint distribution of counterfactual outcome surrogate pairs is not identifiable. We propose a distributionally robust framework for safe and generalizable surrogate evaluation that explicitly accounts for this uncertainty. By treating cross arm dependence as partially identified, we formulate a maximin optimization problem that learns an optimal surrogate transformation by maximizing the worst case explained variance of individual treatment effects over admissible counterfactual dependence structures. The resulting robust surrogacy measure provides a conservative lower bound on explained treatment effect variation. Moreover, our maximin construction guarantees nonnegative alignment between the treatment effects on the transformed surrogate and the primary outcome over the admissible counterfactual couplings. To generalize surrogacy evaluation across populations under covariate shift, we integrate transfer learning with debiased machine learning to estimate the required target population moments using source trial data and target baseline covariates. The computational tractability and asymptotic properties of our method are established. Simulations demonstrate good finite sample performance and show that the proposed method guards against counterfactual misalignment by shrinking the learned transformation toward zero, revealing a robustness prediction tradeoff. We illustrate the framework by evaluating 18 month glomerular filtration rate change as a surrogate for two year change using multiple randomized clinical trial data sets.
Renjie Peng, Christian Genest, David A. Stephens
Tue 17 Nov
stat.COmath.STstat.MLstat.TH
pdf html
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.79
- interest Computational methods for heterogeneous agent and dynamic economic models 0.76
- reading summary, intractable, inference, actually
Abstract
Simulation-based inference (SBI) enables posterior estimation when the likelihood is intractable but data can be simulated from the model. For high-dimensional or variable-length data, SBI methods typically rely on low-dimensional summary statistics as conditioning inputs for posterior estimation, making their sufficiency crucial for accurate inference. Mutual information (MI) provides a natural information-theoretic characterization of sufficiency, quantifying how much parameter information a summary retains. However, these MI quantities are generally intractable in likelihood-free settings. Sui et al. (2026) estimate MI from a trained neural posterior using the Barber--Agakov (BA) lower bound, while discussing, but not computing, a Fisher-information upper bound. This makes the estimate ambiguous: a small lower bound may indicate either an insufficient summary or an inaccurate density estimator. We address this limitation with a tractable diagnostic that estimates the BA gap and reveals what the difference between bounds has actually measured. We extend the resulting sufficiency metric to observation-centered restricted data spaces, allowing simulations to focus on a neighborhood of the data while preserving the target posterior and enabling valid sufficiency comparisons between summaries. We illustrate the use of this diagnostic tool on a temporal epidemic-type aftershock sequence model by comparing learned and hand-crafted summary statistics for variable-length earthquake catalogs.
Konstantinos Grammenos, Dionissios Hristopulos
Fri 13 Nov
stat.OTcond-mat.dis-nnq-bio.NC
pdf html
- interest Computational methods for heterogeneous agent and dynamic economic models 0.77
- reading dimensionality, spectrum, information processing, effective dimensionality
- favorites near “Why Does Deep Learning Work? - A Perspective From Group Theory” 0.79
Abstract
Criticality in neural networks is crucial for optimal information processing. Recent studies have shifted focus from static criticality to frequency-dependent behavior, particularly in oscillatory systems modeled by block-Gaussian orthogonal ensemble (block-GOE) matrices. However, analytical tools to predict spectral properties such as power spectrum and effective dimensionality remain scarce. In this work, we derive exact solutions for the network-averaged power spectrum $\operatorname{Tr}_{N}[\mathbf{S}(\omega)]$, the network autocovariance $\operatorname{Tr}_{N}[\mathbf{C}(\tau)]$, and the normalized effective dimensionality $D_N(\omega)$ in block-GOE ensembles. These results provide a comprehensive framework for understanding how oscillatory networks operate near criticality, offering insights into the dynamic balance between stability and flexibility in large-scale neural systems.
Liyuan Mo, Qingyuan Zhao
Mon 16 Nov
stat.ME
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.85
- interest Causal inference and identification in econometrics 0.80
- reading causal, causal effects, instruments, feedback
Abstract
Mendelian randomization (MR) uses genetic variants as instruments to estimate causal effects. Bidirectional MR usually uses two reversed unidirectional analyses, but feedback can invalidate instruments in at least one direction and induce correlated pleiotropy. We model both traits as a stable cyclic system and estimate both directional effects from GWAS summary data without instrument selection or a prespecified valid-instrument subset. After LD score normalisation, marginal associations form a heteroscedastic noisy independent component analysis model, assuming the variants' direct effects on the two traits are independent. If those effects are non-Gaussian, this identifies the total-effect matrix up to scaling and permutation; no-self-feedback and stability of the system further resolve these ambiguities. We develop an estimator based on noise-adjusted second- and fourth-order moments, avoiding inversion of the high-dimensional linkage disequilibrium matrix and having computational cost linear in the number of variants. Under weak local dependence, we establish consistency, asymptotic normality, and validity of a chromosome-respecting moving block bootstrap. Simulations show negligible bias, generally reliable interval coverage, and higher rejection rates for nonzero directional effects than focusing-based tests. Three real-data applications illustrate how joint analysis of feedback systems can yield causal decompositions differing from two separate unidirectional analyses.
Jake Koerner, Ana-Maria Staicu, Caitrin Murphy, Eric Laber
Tue 17 Nov
stat.ME
pdf html
- reading functional, functional data, eigenfunctions, functional observations
- your work near “Efficient Online Estimation of Causal Effects by Deciding What to Observe” 0.79
- interest Semiparametric inference and debiased machine learning 0.82
Abstract
Traditional functional principal component analysis (FPCA) is typically designed for continuous functional observations. In this paper, we address the scenario where the outcome consists of repeated categorical data defined over a bounded interval. Our objective is to develop an FPCA methodology tailored specifically for ordinal functional data. Our approach leverages recent advancements in ordinal data modeling to estimate both the mean function and the eigenfunctions, while employing a computationally efficient method for predicting the scores. The performance of the proposed methodology is evaluated numerically in simulation and data application on diabetes intervention.
Bo-Yi Lin, Michka Sharpe
Fri 13 Nov
stat.AP
pdf html
- reading state, entry, advances, markov
- your work near “Timing as an Action: Learning When to Observe and Act” 0.75
- interest Bayesian computation and structural estimation 0.81
Abstract
Detecting a wet-weather regime and deciding when to replace slick tyres are distinct tasks. This study examines their relationship through a retrospective analysis of the 2024 British Grand Prix. A three-state hidden Markov model (HMM) represents normalized lap time and precipitation, while an entry rule requires both a high filtered Wet-state probability and sufficient pace deterioration. The analysis includes 642 car-laps and 33 complete lap-bin observations within the available precipitation record. Each completed-lap signal takes effect on the following lap. The combined rule enters on lap 27, matching the team-weighted modal reference; removing the pace or state condition advances entry to laps 25 and 20, respectively. The combined entry is retained in all 10 team-exclusion refits, but refitting a specification with larger off-diagonal transition probabilities produces entry on lap 20. These comparisons illustrate how state inference can contextualize a performance criterion while exposing dependence on model assumptions. Because estimation and rule development use the same event, the correspondence establishes neither prospective accuracy nor a race-time advantage.
Xinyi He, Jianchang Lin, Rachael Liu, Ying Yuan
Tue 17 Nov
stat.ME
pdf html
- favorites near “Bayesian Semiparametric Model for Sequential Treatment Decisions with Informative Timing” 0.86
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.77
- reading success, evidence, covariate shift, conventional
Abstract
Probability of success (PoS) is widely used to inform go/no-go decisions for phase III trials based on phase II evidence. Conventional PoS calculations, however, implicitly assume that baseline covariate distributions are identical across phases. When covariate shift exists between phase II and phase III populations, unadjusted PoS may be biased and lead to miscalibrated decision-making. We propose a Bayesian framework for PoS evaluation that explicitly accounts for covariate shift between phase II and III population prior to phase III initiation. The method adopts a transportability perspective, adjusting phase II evidence to reflect an anticipated phase III target population using available target covariate information. The framework accommodates both individual-level and aggregate-level phase III covariate data. Simulation studies demonstrate that, compared with the conventional approach, the proposed method improves PoS accuracy and decision calibration when covariate shift is present, while maintaining comparable performance when populations are aligned. We illustrate the proposed methodology using clinical trial data from systemic lupus erythematosus.
Antoine Faul, David Ginsbourger, Ben Spycher, Petra Stute
Wed 18 Nov
stat.AP
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.83
- favorites near “Sequential Monte Carlo Steering of Large Language Models using Probabilistic Programs” 0.76
- interest Language models as statistical objects 0.75
Abstract
The generation of synthetic data offers a promising way to address challenges related to data scarcity and privacy concerns. While many approaches have been developed, tabular synthetic data generation methods often require complex models and extensive training data, while lacking interpretability. In this work, we investigate versatile probabilistic approaches using copulas, and leverage in particular the Gaussian Mixture Copula Model, to model the joint distribution of multivariate data, with a main focus on moderate-dimensional data sets. Taking a use-inspired perspective, we examine key trade-offs inherent to synthetic data evaluation, spanning fidelity, utility, and privacy, and demonstrate that copula-based methods achieve a sound balance across all three dimensions while remaining interpretable and providing an explicit probabilistic model. We illustrate and highlight the potential benefits of the method on a test case focusing on systematic missing values in a cardiovascular risk calculator and develop specific evaluation metrics for this application.
Juraj Marusic, Jonathan Wenger, Beau Coker, John P. Cunningham
Fri 13 Nov
cs.LGstat.ML
pdf html
- interest Computational methods for heterogeneous agent and dynamic economic models 0.76
- favorites near “Fit without fear: remarkable mathematical phenomena of deep learning through the prism of interpolation” 0.79
- reading variational, theory, implicit, generalize
Abstract
Variational approaches in deep learning promise to deliver improved uncertainty quantification and out-of-distribution generalization by inferring a distribution over weights, regularized by a divergence to a chosen prior. However, prior elicitation presents a significant challenge in deep learning, resulting in the common practice of significantly downweighting the regularization term in the variational objective. Recently, it was proposed to train solely via the expected loss by relying on implicit regularization from the optimizer, rather than explicit regularization from the divergence to the prior. While this approach largely sidesteps the issue of prior elicitation, current theory lacks an understanding of how key optimization hyperparameters impact the implicit regularization. Here, we theoretically and empirically demonstrate that larger learning rates and fewer parameter samples lead to flatter minima of the loss landscape, which are known to generalize better. We show that training on the expected loss operates at a modified edge of stability, characterized by the signal-to-noise ratio of the gradient estimates and the number of Monte Carlo samples. Experiments on out-of-distribution benchmark datasets confirm the practical significance of the theoretical results.
Hubert Strauss, Liam Fowl, Noam Razin
Thu 19 Nov
cs.LGcs.AIstat.ML
pdf html
- favorites near “Sequential Monte Carlo Steering of Large Language Models using Probabilistic Programs” 0.80
- reading language model, choice, language, far
- interest Language models as statistical objects 0.76
Abstract
On-policy distillation (OPD) trains a student language model to match a teacher model based on its own generations. When using OPD, one must choose whether to apply it at the sequence or token level and which divergence to minimize. However, despite the wide adoption of OPD in practice, the formal understanding of which OPD objective is appropriate for a given setting is limited. Toward addressing this gap, we theoretically analyze how the choice of OPD objective governs which tokens are prioritized, in what order, and how efficiently. Our analysis reveals that sequence-level OPD exhibits a future agreement bias, which myopic token-level objectives avoid. This bias induces a reverse learning order, in which sequence-level OPD first matches the teacher at later token positions, while at earlier positions it can favor tokens that are improbable under the teacher, and thus often undesirable. As for the choice of divergence, we prove that it strongly influences the amenability of OPD to gradient-based optimization. In particular, while the forward KL divergence enjoys a well-behaved objective landscape, other $f$-divergences, such as reverse KL, can suffer from vanishing gradients even when the student and teacher are far apart. Across language model post-training settings, we demonstrate the implications of these theoretical findings for which OPD objective is preferable, highlighting the importance of accounting for the interplay between the student, teacher, and task at hand.
Fabian Dunker, Jyotsna Garg
Fri 13 Nov
econ.EMstat.AP
pdf html
- reading policy, market, standard errors, incentives
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.73
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.68
Abstract
New Zealand's Clean Car Discount (CCD) was a feebate scheme that offered rebates for low-emission vehicles from July 2021, added fees for high-emission vehicles from April 2022, and was abolished at the end of December 2023. Shortly afterwards, road user charges were introduced for electric vehicles (EVs). We study how EV adoption responded to the introduction and the abrupt repeal of the scheme, using monthly vehicle registration data from 2014 to 2024. A structural break analysis following Bai and Perron, combined with a regression discontinuity in time, shows that breaks in the registrations of battery electric, plug-in hybrid and conventional vehicles coincide with the policy dates. To estimate causal effects, we apply a difference-in-differences approach with Queensland, Australia, as a comparison region and Newey-West standard errors. Because the New Zealand market was about four times larger, the analysis uses log-transformed registrations, so that parallel trends refer to equal growth rates. The results are asymmetric. After the introduction of the rebates, EV registrations in New Zealand rose strongly, but by a similar proportion as in Queensland, and the average effect relative to Queensland is not statistically significant. A specification allowing for a change in slope suggests a gradual acceleration of EV uptake. After the repeal, EV registrations fell by about 72% relative to Queensland, and the decline persisted in 2025. Both policy dates were preceded by a rush of purchases. The average CO2 emissions of newly registered vehicles fell markedly during the scheme and rebounded after its end. The results suggest that an abrupt end of EV purchase incentives can cause a large and lasting drop in EV uptake, and that a gradual, pre-announced phase-out may be preferable.
Sie Hendrata Dharmawan, Peter Chin
Thu 19 Nov
stat.MLcs.LGmath.STstat.TH
pdf html
- your work near “Automated Solution of Heterogeneous Agent Models” 0.79
- interest Computational methods for heterogeneous agent and dynamic economic models 0.76
- favorites near “Introduction to regularity structures” 0.82
Abstract
A graphon is determined by all of its homomorphism densities, but motif-based methods use finitely many. We compute exactly what that truncation costs in a natural family: latent structure block models whose positions lie on $K$ rays, equivalently degree-corrected block models of rank two. For a given radial law, the configurations that agree on every graph with fewer than $2K-1$ vertices form a one-dimensional fiber, which we give in closed form with explicit endpoints; the friendship graph $F_{K-1}$ is its unique edge-minimal separator, so no fixed motif order serves every $K$. The separating signal is suppressed by a factor of at most $m_1^{2(K-1)}$, where $m_1$ is the mean squared slope, about $10^{-4}$ at $K=4$: detecting it from motif counts would take about $10^{15}$ nodes in a single graph, at any edge density. The information is nonetheless in the graph; the fiber belongs to the statistic, not the model. Common spectral pipelines, which count no motifs, still slide along the fiber by more than a third of its length at ordinary densities, with confidence regions that almost never contain the truth, while model-based fits of the same embedding do not. We give practitioners four checks: whether a fit is at risk, what to run instead, why a confidence interval will not warn, and how wide the remaining ambiguity is.
Daniel Kua, Yan Song
Tue 17 Nov
stat.MLcs.LG
pdf html
- favorites near “Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss” 0.82
- your work near “Valid Inference with Imperfect Synthetic Data” 0.78
- interest Language models as statistical objects 0.79
Abstract
Generative data augmentation adds samples from generative models to a classifier's training set. Many influential studies use generators pre-trained on large external datasets, and so mix the value of synthetic data with knowledge transferred through pretraining. We study the case without external data, where each generator learns only from the classifier's own training set, and ask when and why adding generated samples to that training set raises or lowers the classifier's test accuracy. We train classifiers with four backbones on six datasets, two of them medical, adding samples from four generators. We bound the excess risk of a classifier trained with synthetic data, that is, how far its population risk can lie above the minimum risk within its hypothesis class. The bound charges synthetic data for the total variation distance between the real and generated distributions and credits synthetic data for a larger training set. The gain from adding synthetic data, the fall in population risk, is at most the excess risk of the real-only classifier, which trains on the real images alone. Read through the bound, the experiments suggest that synthetic data help most when our classifiers detect only a small discrepancy between the generated and real data and the real-only classifier has room to improve, and that synthetic data help little or hurt when the detected discrepancy is large. Generators with a larger lower bound on the total variation distance, a bound that we estimate from our own classifiers, tend to help less on five of the six datasets, and on the CIFAR datasets the most accurate backbone gains the least. The bound gives these directions but not the size of the largest gains, most of which traditional augmentation also reaches at the same number of optimizer steps.
Kevin Rojas, Felix X.-F. Ye, Molei Tao
Thu 19 Nov
cs.LGstat.ML
pdf html
- favorites near “Sequential Monte Carlo Steering of Large Language Models using Probabilistic Programs” 0.82
- reading state, instead, handles, spaces
- interest Language models as statistical objects 0.75
Abstract
Diffusion and flow-based generative models have achieved remarkable results across diverse applications, but most assume data of fixed-dimension/length. Meanwhile, real-world problems such as layout generation often involve variable-length, multimodal data, such as continuous bounding box coordinates paired with discrete types. Existing methods typically address variable length by padding data to a fixed size or by introducing specialized objectives to model lengths explicitly, adding complexity and limiting generality. We propose Variable Length Diffuse Everything (VL-DE), a simple and general framework for variable-length multimodal generation on arbitrary state spaces. Instead of padding, VL-DE handles variable length through an insertion-based mechanism grounded in a principled theoretical framework, yielding a tractable training objective. In addition, by utilizing the fact that VL-DE is based on a generator with an additive form, we develop a higher-order sampler by appropriately handling the time dependence. Extensive experiments on layout and molecule generation demonstrate the effectiveness of our approach.
Peida Li
Tue 17 Nov
cs.LGstat.ML
pdf html
- favorites near “Spectral State Space Models” 0.76
- interest Semiparametric inference and debiased machine learning 0.84
- your work near “Timing as an Action: Learning When to Observe and Act” 0.76
Abstract
We study prediction of nonlinear audit functionals from scalar observations of a shared hidden table state. For Schur-rational operators, local Gaussian projections yield a common sparse kernel for overlapping tasks, repeated row identities, and cross-row dependence. Shared-state residual samples extend this kernel through a positive-semidefinite feature representation. Positive matrix integration and sample-value enclosures connect the original inputs to the actual saved hybrid prediction. A sorted calibration-role perturbation bound then certifies its complete finite-pool pinball risk. Independent complete-loss selection chooses the ridge, score scale, and residual weight, while task-rank calibration supplies an upper endpoint. Evaluation uses twenty new reference contexts per dataset at two task sizes, giving 400 conditions across ten benchmark populations. The mean dataset complete-loss ratios are 0.9977 to the degree-two kernel, 0.9961 to matched complete-MC shrinkage, and 0.9911 to the retrospective minimum over 174 public configurations. The last ratio has a paired reference-bootstrap interval of [0.9898,0.9927]. On 40 profiles at coefficient tolerance $10^{-7}$, Gamma quadrature has a geometric mean cold-time ratio of .0646 to strengthened Arb/Petras integration.
Hung Hung, Jia-Rou Liu
Mon 16 Nov
stat.ME
pdf html
- interest Semiparametric inference and debiased machine learning 0.86
- your work near “Perturbation Methods for Incomplete Markets Economies” 0.77
- favorites near “Variational Bayes under Model Misspecification” 0.85
Abstract
Canonical correlation analysis (CCA) is a widely used statistical tool for exploring the dependence structure between two sets of variables. An important issue in CCA is determining the rank. Among information-based criteria for rank selection, the Akaike information criterion (AIC) and the Bayesian information criterion (BIC) are two commonly used approaches. Both rely on the number of free parameters to quantify model complexity, which may not adequately account for the underlying dependence structure when the assumed model is misspecified. As a result, AIC and BIC may exhibit unsatisfactory performance, particularly when the remaining canonical correlations do not follow the assumed low-dimensional structure. To alleviate this difficulty, we adopt the idea of the generalized information criterion (GIC) to develop a data-adaptive measure of model complexity for CCA rank selection. The proposed complexity measure incorporates the magnitudes of the sample canonical correlations, thereby accounting for the spectral structure of the data and improving robustness to model misspecification. We investigate the asymptotic properties of the proposed GIC under a high-dimensional setting in which the sample size and both variable dimensions grow proportionally. Our theoretical analysis aims to characterize the conditions under which GIC can distinguish signal canonical correlations from noise and achieve more reliable rank selection than conventional information criteria.
Edna C. Cieslik, Lennart Frahm, Felix Hoffstaedter, Robert Langner, Changyi Kuang, Peter T. Fox, et al.
Wed 18 Nov
stat.AP
pdf html
- reading toolbox, conceptual, straightforward, functional
- interest Empirical Bayes, shrinkage and decision theory 0.81
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.75
Abstract
Coordinate-based neuroimaging meta-analysis using activation likelihood estimation (ALE) is commonly used to synthesize reported results of functional or structural magnetic resonance imaging studies. We here present JALE, a new, open-source Python package for conducting ALE meta-analyses. Besides computing standard ALE main effects and contrast analyses, this toolbox implements several novel analytical approaches including new options for thresholding (threshold-free cluster enhancement statistics, significance threshold prediction) as well as sensitivity and cluster analyses that were only available as in-house scripts in the past. The purpose of this paper is to provide an integrated conceptual foundation of recent developments in the ALE meta-analysis framework, with a particular focus on novel methodological implementations that can be executed in a straightforward and standardized manner using the new JALE toolbox.
Ergan Shang, Si Chen, Canhong Wen
Mon 16 Nov
stat.MEmath.STstat.MLstat.TH
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.78
- interest Computational methods for heterogeneous agent and dynamic economic models 0.77
- favorites near “Finding Regularized Competitive Equilibria of Heterogeneous Agent Macroeconomic Models via Reinforcement Learning” 0.76
Abstract
Data in forms of networks has been encountered in various prevalent fields at the arrival of big data era. To unveil and recover the true model with core units with numerous links, the $\beta$-model under sparsity constraint is proposed to meet the expectation of statisticians and engineers. Not only does this network model inherit spectacular property from the Erd\H{o}s-R\'enyi model, but also exhibits degree heterogeneity through tuning sparsity parameter $s$. In this paper, we propose Homogeneous Pursuit (HP), an $\ell_0$-constrained maximum likelihood estimation method implemented through a primal-dual algorithm. To ease the harshness of computation, we find the surrogate function of the corresponding objective function to derive Majorization Minimization algorithm through the primal-dual process. Moreover, after meticulous evaluation, the algorithm brought about captures the property of fine convergence rate in a contraction mapping way. Under some mild conditions, the proposed generalized information criterion (GIC$^*$) for selecting the sparsity level is consistent to find the truth. Combining HP with GIC$^*$ gives the adaptive procedure HP-GIC$^*$; HP-BIC denotes the same estimation method with sparsity selected by BIC. Finally, simulation studies show that GIC$^*$ improves support recovery relative to BIC in the reported comparisons, while HP achieves lower computational cost while maintaining estimation accuracy comparable to the sparse benchmark methods. A microfinance application illustrates the use of HP-GIC$^*$ for studying the association between network prominence and participation, with improved fit in the reported village-level regressions.
Simon Hirsch, David Schaurecker, David Wozabal
Fri 13 Nov
q-fin.STq-fin.CPq-fin.TRstat.AP
pdf html
- favorites near “Eight Centuries of Global Real Interest Rates, R-G, and the 'Suprasecular' Decline, 1311-2018” 0.74
- reading markets, market, price, burden
- interest Computational methods for heterogeneous agent and dynamic economic models 0.73
Abstract
Grid-scale battery energy storage systems (BESS) are becoming increasingly important in short-term energy markets, where they help balance fluctuations in variable renewable generation. In many countries, the continuous intraday market is the most important wholesale market for BESS to monetize their flexibility. In this paper, we propose an extension to the myopic rolling intrinsic policy that incorporates intraday price forecasts tailored specifically to the BESS trading problem. While a large literature addresses intraday price forecasting, most existing work focuses on volume-weighted prices during the last hours of trading and consequently fails to capture the features most relevant to BESS trading. Our approach addresses this gap by introducing a novel target variable derived from the limit order book and its dynamics. The resulting forecasts are designed to inform the optimization about future profit opportunities while preserving computational efficiency and avoiding the burden of full stochastic optimization. In a realistic high-frequency simulation setting covering 2023 and 2024, our method improves profits by up to 42% over the naive rolling intrinsic formulation. We further analyze the relationship between forecast accuracy, stability, and financial performance, establishing a clear inverse relationship between accuracy and stability. Finally, we find evidence that BESS size and duration jointly influence both overall profitability and the marginal benefit of incorporating forecasts.
Qisheng Wang
Wed 18 Nov
cs.ITmath.ITmath.STstat.TH
pdf html
- reading hilbert space, hilbert, short, proofs
- interest Adaptive data collection and anytime-valid inference 0.81
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.72
Abstract
Min-entropy plays a fundamental role in assessing the unpredictability of randomness sources. In particular, NIST SP 800-90B specifies min-entropy estimation as a key procedure for entropy sources used in cryptographic random bit generators, illustrating the practical importance of reliable min-entropy estimation. In this paper, we settle the sample complexity of min-entropy estimation with high probability in both the classical and quantum cases. Specifically, for min-entropy estimation with additive error $\varepsilon$ and probability at least $1-\delta$, the sample complexity is $\Theta(d\log(d/\delta)/\varepsilon^2)$ for classical distributions on $d$ symbols and $\Theta((d^2+d\log(1/\delta))/\varepsilon^2)$ for quantum states on a $d$-dimensional Hilbert space. Our results admit short and simple proofs.
Mallikarjuna Rettiganti, Yipeng Wang, Yongming Qu
Wed 18 Nov
stat.ME
pdf html
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.85
- reading tipping, treatment, versus, mg
- interest Causal inference and identification in econometrics 0.76
Abstract
Missing data remain a persistent challenge in clinical trials despite preventive efforts. Handling missing data in clinical trials requires not only a primary analysis but also sensitivity analyses, as the underlying missingness assumptions are difficult to verify. The {\delta}-adjusted analysis, which assesses the robustness of primary results by adding a sensitivity parameter {\delta} for one or both treatment groups, has been widely used for continuous outcomes. Then, a set of tipping points which flip the significance of the primary analysis can be identified. However, there is no corresponding {\delta}-adjusted method for binary variables. In this article, we propose a {\delta}-adjusted method for binary variables that addresses this gap. This method assesses robustness by adding a sensitivity parameter to the logit function of the estimated response probability under the primary imputation approach and shares the same property as the {\delta}-adjusted method for continuous variables. We applied this method to a clinical trial evaluating the treatment effect of tirzepatide 15 mg versus 5 mg on a composite binary endpoint. This method provides a principled sensitivity analysis framework for binary endpoints with broad applicability in clinical trials.
Tiandong Wang
Fri 13 Nov
math.STstat.MEstat.TH
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.79
- favorites near “Random Walks on Simplicial Complexes and Harmonics” 0.82
- interest Semiparametric inference and debiased machine learning 0.82
Abstract
Minimum-distance fitting selects a degree threshold and a power-law exponent from the same network. We study the procedure that fits a continuous Pareto law with the Hill estimator and minimizes the Kolmogorov--Smirnov distance over retained ranks. For homogeneous linear preferential attachment and Norros--Reittu graphs with Pareto weights, we derive the joint limiting law of the selected rank, retained fraction of the threshold tie, and exponent error. The selector retains of order $n^{2/(\alpha+2)}$ of $n$ vertices, where $\alpha$ is the survival exponent. At this rank, integer spacing and finite-degree departures from a Pareto tail have the same order as sampling fluctuations. The exponent estimate is consistent and retains a shift on the scale used for confidence intervals. For integer degree laws whose tail sums have a first-order inverse-degree expansion, adding the retained tie fraction divided by the threshold to the Hill statistic removes the leading contribution of the tie. Under homogeneous preferential attachment with a fixed known number of edges per new vertex, removing the remaining model-specific term and calibrating for selection gives confidence intervals with asymptotically exact coverage. Observed networks show sensitivity of the fitted exponent and the classification of degree variance to threshold ties. Simulations assess interval coverage and show improved pooled connectivity predictions after correction under both specified graph laws. Changes to the Norros--Reittu weight distribution can reverse the prediction gain.
Abbas Raza Ali, Muhammad Ajmal Siddiqui, Moona Zahid
Thu 19 Nov
cs.AIcs.MAstat.ML
pdf html
- reading state, lessons, business, bank
- favorites near “Data Science at the Singularity” 0.82
- your work near “Perturbation Methods for Incomplete Markets Economies” 0.75
Abstract
Agentic AI platforms-systems in which LLM agents plan, invoke tools, retain memory and coordinate to change the state of enterprise systems-are being procured faster than they can be measured. Public agent benchmarks score models in fixed harnesses; enterprises buy integrated platforms whose orchestration, memory, governance and tooling layers dominate outcomes, and they buy them largely on the strength of vendor demonstrations. We report an industrial evaluation programme ("bakeoff") run by the innovation laboratory of a global bank in which ten platforms-spanning open-source frameworks, proprietary products and hyperscaler ecosystems-were assessed by a cross-functional panel on nine capability dimensions across three financial-services workloads: counterparty credit risk, retrieval-augmented generation and process automation. Scores separate the field into three tiers that are stable under alternative weightings, and surface workload-specific strengths that aggregates hide. A secondary analysis, however, finds Cronbach's alpha = 0.98 across nominally distinct dimensions and rank correlations >= 0.93 with the total for eight of nine: a halo effect indicating that demonstration-based scoring measures presentation as much as capability. From these lessons we propose a gate-then-score evaluation protocol with four layers (capability, execution reliability, governance gates, business value), trace-level reliability metrics estimated over repeated runs, an autonomy-readiness scale and a reproducibility protocol. We state precisely what the present evidence supports, and what a controlled harness must add to turn enterprise platform evaluation from structured judgement into measurement.
Amirhosein Farahmand, Mauricio Lopez-Mendez, Mohsen Sadatsafavi, Abdollah Safari, Tae Yoon Lee
Wed 18 Nov
stat.ME
pdf html
- favorites near “Bayesian Semiparametric Model for Sequential Treatment Decisions with Informative Timing” 0.85
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.80
- reading nominal, net, conventional, specifications
Abstract
Decision curve analysis (DCA) is recommended for evaluating the clinical utility of prediction models by quantifying net benefit (NB). Conventional NB estimators for survival outcomes rely on the assumption that censoring is non-informative, potentially yielding biased estimates and misleading decisions when this assumption is violated. We propose a robust estimator of survival NB using conditional inverse probability of censoring weighting (CIPCW). In simulation studies, we compared its performance against the conventional approach across censoring mechanisms, censoring model specifications, censoring rates, and decision thresholds, in terms of relative bias, relative root mean squared error (RMSE), and 90% confidence interval coverage. Under non-informative censoring, both estimators performed comparably, with relative bias below 1% and coverage close to the nominal level. Under informative censoring, conventional estimators underestimated the true NB by up to 58% with coverage as low as 1%, whereas the CIPCW estimator had a relative bias of at most 4%, coverage of at least 78% and consistently lower relative RMSE. As a case study, we illustrated its use in validating a breast cancer recurrence-free survival model to guide intensified adjuvant therapy in the German Breast Cancer Study Group (GBSG) cohort. Using the model was the best strategy at some clinically relevant thresholds, but accounting for potential informative censoring changed the best strategy to treating all patients, showing that informative censoring can change the conclusions drawn from DCA. When censoring may be informative, we recommend CIPCW-based NB estimation in DCA for survival outcomes.
Nischal Subedi, Cencheng Shen, Peng Zhao
Thu 19 Nov
cs.LGstat.ML
pdf html
- favorites near “Spectral State Space Models” 0.79
- reading gains, theory, explains, view
- interest Language models as statistical objects 0.78
Abstract
We show that the importance of a LoRA adapter's initial input-subspace direction is regime-dependent, and we characterize this dependence in theory and experiment. This refines the recent view that spectral initializers help through update magnitude alone. Direction matters most when labeled data are scarce relative to the representation dimension and the activation subspace used for initialization captures supervised task structure. In this regime, the adapter's row span remains sensitive to its starting point, making task-aligned initialization especially useful. A spiked-covariance perturbation analysis characterizes the finite-sample stability of the estimated subspace. A first-step analysis of LoRA training explains why the subspace's alignment with the task matters and motivates an alignment coefficient $\rho$ that closely tracks gains across classification tasks. We test this account with PivotLoRA, which initializes adapters from activation PCA at prediction-relevant positions and applies one output-preserving orthogonal pivot early in training. In few-shot BERT-base classification, PivotLoRA achieves the best average rank among 13 methods and improves over vanilla LoRA in 29 of 30 settings, even though it starts from a smaller input-factor initialization norm. Additional classification experiments extend the evaluation to 7B and 8B Llama decoders. Classification gains diminish with more labeled data, consistent with the proposed regime dependence. A targeted extension to completion-only reasoning also shows improvements on GSM8K and StrategyQA. These findings identify conditions under which initialization direction matters, with subspace-task alignment offering a low-cost diagnostic before training for classification.
Zilin Wang, Binghui Li, Jia-Nan Wang, Lean Wang, Jinbo Wang, Lei Wu
Fri 13 Nov
stat.MLcs.LG
pdf html
- interest Kernels, operators and functional data 0.83
- favorites near “Recent Developments in Machine Learning Methods for Stochastic Control and Games” 0.77
- your work near “Computational Methods for Economic Models with Function Valued States” 0.76
Abstract
We study how the sign operator interacts with gradient noise to shape the learning dynamics and scaling behavior of sign-based optimizers. In power-law kernel regression with diagonal features, we establish a functional scaling law (FSL) for SignSGD under general learning rate schedules. The FSL describes the learning dynamics through two processes: signal learning and noise forgetting. The key mechanism is noise-induced preconditioning: gradient noise smooths the sign operator, inducing an effective preconditioned SGD. Noise-curvature alignment enables this preconditioning to transform the effective Hessian from $\mathbf{H}$ to $\mathbf{H}^{1/2}$. The resulting dynamics exhibit faster signal learning but slower noise forgetting than SGD. This tradeoff yields a regime-dependent comparison with SGD: (i) SignSGD achieves improved data efficiency in the hard-task regime; (ii) both optimizers achieve the same data efficiency in the easy-task regime, but SignSGD maintains this efficiency at larger batch sizes, requiring fewer optimization steps for a fixed data budget. Power-law kernel regression experiments support the proposed mechanism and predicted scaling behavior. LLM pre-training experiments further suggest that the SignSGD FSL better captures the batch-size dependence of loss dynamics for sign-based optimizers than the original SGD FSL.
Yidong Zhou, Muqing Cui, Hans-Georg Müller
Fri 13 Nov
stat.MEcs.AI
pdf html
- favorites near “Monte Carlo Geometry Processing” 0.81
- interest Kernels, operators and functional data 0.79
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.77
Abstract
Scientific observations such as probability distributions, trees and covariance matrices call for generative models that respect their non-Euclidean geometry. While flow matching has been extended to manifolds by utilizing geodesics and velocities, such manifold structures are not available in a general metric space. We introduce Fr\'echetFlow, a generative modeling framework for metric space-valued data that requires neither linear operations nor a smooth manifold structure. A key feature of the proposed approach is that we utilize well-established Euclidean flows to learn the joint distribution of distances to a set of landmarks, whereupon new objects are reconstructed from generated landmark representations. The reconstruction step utilizes weighted Fr\'echet means of reference objects, drawing on Fr\'echet regression. We establish Wasserstein error bounds that connect accuracy of the generated object distributions to the placement of landmarks, flow approximation and reconstruction error and prove consistency under finite second moments. This leads to convergence rates that depend on the metric complexity of the target objects, tail moments and fitting accuracy. Experiments show competitive accuracy with established flow matching methods in Euclidean simulations and lower Wasserstein errors than specifically tailored methods for several scientific datasets. Across simulation settings, we find that generation error decreases as sample size increases.
Bisheshwor Neupane
Fri 13 Nov
stat.AP
pdf html
- reading household, forecasting, did, consumption
- interest Computational methods for heterogeneous agent and dynamic economic models 0.74
- your work near “Automated Solution of Heterogeneous Agent Models” 0.76
Abstract
This study benchmarks 24 forecasting configurations on ESHED-38, daily smart-meter consumption for 38 homes in Erbil, Iraq, collected from May 2020 to June 2025 with station weather and a household survey. The configurations span five base families, plus two cross-family ensembles: naive baselines, per-house statistical models, global gradient-boosted trees, global deep networks, and zero-shot time-series foundation models. All are evaluated on the same 54,720 forecasts from 48 weekly origins over a 359-day held-out period, at lead times up to 30 days, using scaled absolute error, the continuous ranked probability score, calibration, and Friedman, Nemenyi and Holm-corrected Wilcoxon tests. The best configurations cut the scaled error of a seasonal-naive forecast by 20 to 25 percent, and the leading models beat it in all 38 houses. Chronos-2, used without training, is significantly the most accurate model one day ahead; at one week it is statistically tied with TimesFM-2.5, Chronos-Bolt and a pre-specified ensemble, and the two ensembles lead at 30 days. Deep networks trained from scratch do not outperform exponential smoothing on average, and the household survey adds negligible skill. Metered demand peaks in winter and falls as temperature rises. Recorded weather did not improve any model run with and without it. A quality screen helps explain why: on 68 percent of dates a temperature value recurs exactly on other dates, in three annual blocks covering almost all of the test period, suggesting that much of the weather record reuses one annual sequence. The value of weather information is therefore treated as unresolved, and the protocol is released as a baseline.
Daniel Tesfay, Almaz Abebe
Tue 17 Nov
stat.APmath.DS
pdf html
- reading state, domain-specific, national, strongly
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.75
- favorites near “Robust causal inference with continuous instruments using the local instrumental variable curve” 0.83
Abstract
Medicaid covers nearly 78 million people, yet care quality varies across states, programs, and clinical domains. Using the 2024 CMS Child and Adult Core Sets (9,781 usable state-measure observations from 11,100 records), this study combined descriptive statistics, benchmark-gap analysis, correlation, and OLS regression to find where shortfalls cluster. Disparities were strongly domain-specific. Maternal and Perinatal Health scored lowest (28.0%), followed by Dental and Oral Health (38.5%) and Behavioral Health (44.4%), while Long-Term Services and Supports (82.8%) and Experience of Care (67.8%) scored highest. State performance correlated strongly with national benchmarks (r = 0.77, approximately), so rates should be read against each measure's benchmark. Domain was the strongest predictor, and the model explained 31.2% of variance. Because of heteroscedasticity and influential observations from mixing bounded and utilization measures, estimates are descriptive. Quality improvement should be targeted by domain, prioritizing maternal health, behavioral health network adequacy, and dental provider participation.
Victor Chernozhukov, Denis Chetverikov, Kengo Kato, Yuta Koike
Tue 17 Nov
econ.EMmath.STstat.TH
pdf html
- reading rectangles, little, moment, large
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.80
- interest Adaptive data collection and anytime-valid inference 0.80
Abstract
We study how well a sum of $n$ independent random vectors with $p$ coordinates can be approximated by a Gaussian vector with the same mean and covariance. We measure the error over rectangles, so the bounds apply to maxima and simultaneous confidence regions. The coordinates may be arbitrarily dependent, even exactly linearly dependent, and the bounds do not deteriorate as the covariance matrix becomes singular. After standardizing the coordinate variances, bounded or sub-exponential coordinates give the $n^{-1/2}$ sample-size rate up to logarithmic factors, with the size or tail bounds held fixed. The main upper bounds retain the Gaussian expected maximum, improving the rates when the coordinates are strongly dependent. Fourth-moment refinements give further improvements when large observations contribute little to individual coordinate variances. If we instead impose a fixed moment bound of order $q>4$ on the largest absolute coordinate, together with a coordinate fourth-moment bound, the rate becomes $n^{-1/2+1/q}$ up to logarithmic factors. At $q=4$, the slower $n^{-1/4}$ rate is unavoidable, confirming our earlier conjecture. We also allow the bounds on observation size to grow with $n$. Lower-bound examples show that the dependence on sample size, observation size, and Gaussian width is sharp up to logarithmic factors in the regimes studied, and in several regimes the upper and lower bounds match including the logarithmic terms.
Aarshvi Gajjar, Syamantak Kumar, Aditya Makkar, Christopher Musco, Joey Zhou
Thu 19 Nov
stat.MLcs.LGcs.NAmath.NA
pdf html
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.82
- favorites near “Hierarchies of Relaxations for Online Prediction Problems with Evolving Constraints” 0.80
- interest Online learning and statistical learning theory 0.77
Abstract
In this paper we study active learning for constrained linear regression. Given a design matrix $A\in\mathbb{R}^{n\times d}$, a constraint set $C\subseteq\mathbb{R}^d$, and query access to $b\in\mathbb{R}^n$, we seek an approximate solution to $\min_{x\in C}\|Ax-b\|_2^2$ using as few queries as possible. This problem arises in many scientific and engineering applications, and randomized leverage score sampling provides a powerful method to solve it. However, existing leverage score methods ignore the constraint or encode it through ridge regularization, suffering unfavorable query complexity for anisotropic constraint sets. We propose ellipsoid-regularized leverage scores, where we approximate the constraint set by an outer ellipsoid and use its shape matrix to bias sampling toward directions that matter under the constraint geometry. We prove that the query complexity is governed by an anisotropic effective dimension $d_{\text{eff}}(C)\le d$. With appropriate parameters, $O(d_{\text{eff}}(C)\log d/\varepsilon^2)$ queries suffice to return a point in $C$ with objective value at most $(1+\varepsilon)$ times the optimum. This bound can substantially improve on the standard leverage-score bound, since anisotropic constraints can make $d_{\text{eff}}(C)$ much smaller than $d$. For convex constraints, we improve the dependence on $\varepsilon$ from $1/\varepsilon^2$ to $1/\varepsilon$. We also prove two lower bounds: first, the $d/\varepsilon^2$ dependence is unavoidable without convexity assumptions on $C$, and second, in some regimes, the effective dimension dependence $d_{\mathrm{eff}}(C)/\varepsilon$ is necessary for ellipsoidal constraints. Experiments on synthetic regression problems and a parametric heat conduction problem show lower regression error than standard baselines for similar query budget.
Dario Deniz Ergun, Haohua Wang, Brendon Tao, Jonah Kollenberg, Chen Ye Gan
Thu 19 Nov
cs.AIstat.CO
pdf html
- interest Computational methods for heterogeneous agent and dynamic economic models 0.77
- reading online, large, gives, submission
- your work near “Timing as an Action: Learning When to Observe and Act” 0.75
Abstract
Estimating how an AI system will perform on a new benchmark need not require another large model. We use public evaluation counts to choose a weak Beta prior, then update it exactly from at most 31 revealed outcomes. The predictor uses no item semantics, pretrained weights, or inference API. Its single online evaluation in the AIMS competition returned a team-reported Brier ALC of 0.188477, a 24.6% reduction from the constant-0.5 reference. The complete submission is 7,754 bytes; a server microbenchmark measures 12.71 microseconds per prediction with 31 matching labels. We describe how source evaluations are compressed into 301 prior means and show why a concentration of two lets new evidence rapidly determine the prediction. The resulting posterior gives exact uncertainty about the modelled success rate without numerical integration or neural inference. A public evaluation covers 79 system-benchmark pairs and eight seeds, reaching Brier ALC 0.200250 for the submitted method. Accompanying code and sufficient statistics reconstruct the fitted prior byte for byte and replay all 632 evaluation trajectories. The result demonstrates how much predictive value a compact Bayesian model can extract from sparse target feedback at very low online cost.
Reymart Salcedo Lagunero, Andreas Langer, Cordula Reisch
Tue 17 Nov
stat.MLcs.LG
pdf html
- reading state, good, successful, mechanisms
- interest Computational methods for heterogeneous agent and dynamic economic models 0.74
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.74
Abstract
System identification of nonlinear dynamical systems becomes challenging when the measured data come from a nearly settled state. We study this situation in the case of a bistable host-pathogen interaction model with a cubic term, whose seven true mechanisms are hidden among twenty candidate terms. The data are from one noise-free trajectory on a quasi-steady-state observed window and the initial state. In this situation, classical sparse identification approaches fail to identify the correct mechanisms as the candidate library becomes ill-conditioned along the quasi-steady-state trajectory. To address this, we propose an alternative approach in which sparse Bayesian learning and $\ell_{0}$ iterative hard thresholding with pursuit provide a good start, and a two-swap climb on the Bayesian evidence of the candidate ODE solutions recovers the true model. Our results show successful model recovery, i.e., correct identification of its support and accurate estimation of its coefficients, as validated by exhaustive scoring.
David Jeevankumar, Maryam Sultana, Alexander D. Rast, Fabio Cuzzolin
Fri 13 Nov
stat.APcs.CEcs.LG
pdf html
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.80
- reading counterfactual, choice, functional, literature
- interest Empirical Bayes, shrinkage and decision theory 0.80
Abstract
Machine learning models for mineral exploration are trained on a few hundred sparse, expensive samples and used to decide where to drill next, so their epistemic uncertainty (EU), the part that more data could remove, matters as much as their predictions. Yet EU cannot be measured on real data, because the counterfactual that defines it is never observed, so estimators are graded by proxies such as out-of-distribution detection and calibration. We contribute a general simulation protocol that makes EU directly computable for a given model and can grade any EU estimator. It separates two estimands the literature conflates, reducible error and estimator spread, builds an oracle reference for each in a synthetic prospectivity field sampled by clustered drilling surveys, and compares uncertainty measures of an identical ensemble within each simulated world. We use it to ask when credal (imprecise-probability) measures are more faithful than Bayesian model averaging (BMA) in the sparse, low-class regime of geological classification. The credal interval width recovers the epistemic ordering better than BMA when members are few, confirming the small-ensemble advantage reported for credal wrappers; the advantage vanishes by M = 10 and reverses at M = 40. Secondly, the choice of credal functional matters more than the choice between credal and BMA. Lastly, the credal advantage does not widen as training data are thinned, so it compensates for few posterior samples rather than few data. On real drill-hole geochemistry from South Australia, a resampling oracle reproduces the first finding. The protocol applies to any EU estimator whose data-generating process can be simulated.
Ning Ni, Yongxin Zhu
Mon 16 Nov
stat.MLcs.AIcs.LG
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.78
- reading theory, functional, element, accounted
- interest Bayesian computation and structural estimation 0.79
Abstract
In generative modeling, standard measure theory adopts the a.e. convention, which legitimately ignores any error component carried by a null-support set: a single null element can hide out-of-domain error inside null support and never be explicitly accounted for. The contribution of this paper is not a new distributional distance, but rather a verifiable structure that writes down which underlying predicate supports a high-level object, when it is invertible, and at which layer the error resides, from which we propose the enumerating posterior-supported prior framework. The framework operates on a three-layer categorical chain Quot->MorTop->Set: the first two layers posit no point elements and take binary-predicate quotient partitions as objects; the replica-index counting measure successively discriminates one-to-many collections into injections, lifting object collection from surjection to a quotient-layer bijection; the mutual exclusion of preimages in the shared measure space is resolved by an invertible MPT. Four results consolidate into Thm. 7: binary quotient-partition meta-categorical objects support the set categorical functional along categorical support via measure descent--recovery, enumerating the morphism itself rather than its preimage and object; their multiplicity is normalized through replica-index injection and accounted as an explicitly first-order error. The contribution of this paper is a transformation from nontrivial-input construction -> engineering-computable first-order output: the binary quotient-partition meta-categorical objects, supported via categorical support measure descent--recovery, constitute a substantive nontrivial input---precisely because they lift the out-of-domain error that hid inside null support into a first-order falsifiable coefficient mu_err, a finitely obtainable finite integral first-order error.
Pierre C Bellec
Wed 18 Nov
math.STstat.TH
pdf html
- interest Semiparametric inference and debiased machine learning 0.83
- reading argument, group lasso, lasso, derivatives
- favorites near “Orthogonal Statistical Learning” 0.85
Abstract
For smooth penalties and group Lasso, analyzed approximate leave-one-out (ALO) cross-validation by comparing its weights with the trace correction used in mean-field inference. For a $C^2$ penalty $R$, the matrix entering these corrections is $A=[\nabla^2R(\widehat b)+X^TDX]^{-1}$, where $\widehat b$ is the full-data estimator and $D$ is the diagonal matrix of fitted loss curvatures; for group Lasso, the penalty and loss curvatures are first restricted to the active support before taking the pseudoinverse. The obstacle to extending this argument to general convex penalties was the lack of an explicit formula for the matrix $A$ entering these corrections. For every finite-valued strongly convex penalty, we construct such a matrix from ordinary derivatives of the penalty response at the observed subgradient. Its projection--pseudoinverse formula separates the response space and penalty curvature, fixed by conditioning, from the explicit design dependence. This supplies the missing ingredient to extend the probabilistic argument of Part~I. The construction recovers the inverse Hessian for smooth penalties and the active-group inverse for group Lasso, including inactive boundary groups. In the proportional regime, for Gaussian designs with general positive-definite covariance and covariance-scaled strong convexity, we obtain unconditional bounds comparing both ALO and mean-field predictions with exact leave-one-out predictions in regression and single-index models. The expected average squared prediction error is $O(n^{-1})$, and the expected absolute criterion error is $O(n^{-1/2})$, under bounded, 1-Lipschitz loss scores and suitable test functions. The matrix and ALO weights use only observed data; the mean-field correction additionally uses the design covariance. The construction, recovery identities and statistical guarantees are formalized in Lean~4.
Hongnian Huang, Andrew Stump
Thu 19 Nov
cs.LGcs.CVmath.DGstat.ML
pdf html
- reading manifolds, operators, question, theory
- favorites near “Neural Operator: Learning Maps Between Function Spaces With Applications to PDEs” 0.79
- interest Computational methods for heterogeneous agent and dynamic economic models 0.74
Abstract
Geometric convolutions on Riemannian manifolds depend on the geometry of the underlying domain, making stability under metric perturbations a fundamental question in geometric deep learning. This paper studies this question for parallel transport convolution through methods from differential geometry and analysis. On a fixed compact manifold, we establish that, for a continuous input signal and a continuous reference kernel supported in a sufficiently small normal neighborhood, the convolution output depends continuously in $L^2$ on the Riemannian metric in the $C^2$-topology, under the specified localization and transport-path conventions. The analysis develops perturbation estimates for geodesics, parallel transport, Jacobi fields, and exponential maps, and uses these estimates to compare the transported kernels. A central difficulty is the cut locus, which obstructs uniform comparison of the transport construction over the entire manifold. We address this difficulty by combining uniform estimates away from the cut locus with integral bounds on neighborhoods of arbitrarily small volume, while also controlling the change in the Riemannian volume measure. The resulting qualitative stability theorem connects perturbation theory for Riemannian geometric constructions with the analysis of spatial convolution operators in geometric deep learning.
Irakli Gudavadze
Fri 13 Nov
stat.AP
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.80
- interest Adaptive data collection and anytime-valid inference 0.81
- favorites near “Deep Learning for Individual Heterogeneity” 0.77
Abstract
Sparse count forecasting arises when events are rare, exposure varies across observational units, and short histories contain many zeros. Using thin-market airline demand as a motivating application, this paper develops a hierarchical exposure-adjusted Poisson-Gamma (HEPG) framework for probabilistic forecasting. Conditional counts follow a Poisson model with an exposure offset and pooled calendar effects, while unit specific baseline intensities receive a Gamma prior whose hyperparameters are estimated empirically across the population. The resulting posterior is Gamma and the posterior predictive distribution is negative binomial, providing closed-form partial pooling and explicit propagation of parameter uncertainty. This is especially useful for zero-history or low-exposure units, where independent maximum-likelihood estimates are unstable or collapse to zero. An adaptive extension, A-HEPG, adds a moment-matched negative-binomial variance correction when residual extra-Poisson variation remains. In synthetic experiments with 400 heterogeneous markets, HEPG improves average negative log predictive probability relative to complete pooling and independent route-level Poisson estimation. At a mean latent rate of 0.15 requests per exposure unit, the score is 0.525 for HEPG, compared with 0.585 for pooled Poisson and 0.731 for route-level Poisson maximum likelihood. Under strong overdispersion, A-HEPG reduces the score from 0.552 to 0.529. A downstream cabin-capacity example links the predictive distribution to a seat-protection rule based on predictive survival probabilities and fare ratios. The study emphasizes coherent probabilistic inference, partial pooling, and diagnostic model escalation for sparse applied count data.
Jason K. C. Ng, Thomas B. Berrett, Yi Yu
Thu 19 Nov
math.STstat.MEstat.TH
pdf html
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.81
- interest Adaptive data collection and anytime-valid inference 0.80
- reading differential, moment, kind, literature
Abstract
We study mean estimation under federated differential privacy (FDP) constraints and batch contamination, considering both Gaussian and heavy-tailed distributions. In the homogeneous setting, where sample sizes and privacy budgets are constant across servers, we show that the costs of privacy and contamination disentangle. Under a bounded $v$th moment condition, with $v \in (1, \infty)$, the cost of batch contamination is of order $(\varepsilon/n)^{2-2/v}\vee \varepsilon^2/n$, where $\varepsilon$ is the contamination level and $n$ is the per-batch sample size. This reveals a phase transition at $\varepsilon\asymp n^{1-v/2}$, for $v > 2$, which is the first of its kind in the literature. We develop a two-stage procedure that achieves the minimax rate up to poly-logarithmic factors. A key component is a private smoothed voting procedure tailored to the FDP setting that provides optimal sensitivity control. For heterogeneous batch sample sizes and privacy budgets, we introduce a batch partitioning framework that groups batches with comparable per-batch sample sizes and privacy budgets. Surprisingly, the minimax rate is determined, up to poly-logarithmic factors, by the most informative group of batches. Consequently, in some regimes, it is minimax optimal to use only an arbitrarily small proportion of the data set. Heterogeneity also introduces interactions between the costs of privacy and contamination, in contrast to their disentanglement in the homogeneous setting. For unknown contamination level, we propose an adaptive, near-minimax optimal Lepski-type extension for the heterogeneous Gaussian setting, designed specifically to prevent privacy leakage by ensuring no repeated use of data points.
Jiaye Chen, Rui Qiu, Zhou Yu
Fri 13 Nov
stat.MEmath.STstat.MLstat.TH
pdf html
- favorites near “Distributional Random Forests: Heterogeneity Adjustment and Multivariate Distributional Regression” 0.90
- interest Language models as statistical objects 0.76
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.77
Abstract
We propose Fr\'echet oblique forests (FOF) for non-Euclidean regression with predictors and responses in metric spaces. An Alexandrov inner-product score defines directional splits in metric predictor spaces and recovers ordinary oblique splits for Euclidean predictors. Candidate splits are evaluated through pairwise response distances, avoiding repeated Fr\'echet mean computations during split search. Forest predictions minimize a weighted Fr\'echet criterion obtained by averaging leaf weights across trees. For compatible tree rules, we establish integrated consistency and polynomial risk bounds under explicit approximation and entropy conditions. Simulations with matrices, distributions, spherical data and mixed predictors, together with an application to taxi-flow networks, demonstrate competitive predictive performance.
Yingli Wang, Lingjiong Zhu
Fri 13 Nov
stat.MLcs.LGmath.PR
pdf html
- reading equilibrium, theory, formula, persistence
- interest Computational methods for heterogeneous agent and dynamic economic models 0.75
- favorites near “Recent Developments in Machine Learning Methods for Stochastic Control and Games” 0.75
Abstract
Understanding the persistence of learned behaviors in AI safety motivates an investigation of the dynamical counterpart to the equilibrium geometry studied in singular learning theory. We analyze this persistence through transitions between singular wells. We establish an Eyring-Kramers formula for a low temperature Langevin diffusion between two compact connected zero components of a real-analytic potential. The wells may be singular and nonisolated, and the unique minimizing gate is a nondegenerate index-one saddle, with gates understood as inclusion-minimal sets meeting every minimizing path. A boundary-gate construction shows that uniqueness of this inclusion-minimal gate is equivalent to every minimizing path passing through the saddle and one such path meeting the communication level only there. Compact semianalytic topology provides a separating function away from a fixed saddle neighborhood; analytic Morse coordinates give matching capacity bounds by first sending the temperature to zero and then shrinking that neighborhood. An anchored energy estimate and a small-ball potential-theoretic identity give uniformity for deterministic starting points. The leading constant is expressed in terms of the sublevel-volume coefficient, a Gamma factor, and the saddle Hessian.
Tianyue Qiao, Lijun Sun
Tue 17 Nov
stat.COmath.OCstat.AP
pdf html
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.79
- interest Bayesian computation and structural estimation 0.83
- reading equilibrium, gives, choice, feedback
Abstract
Treating traffic counts as data calls for the joint distribution of link flows. Stochastic user equilibrium (SUE) gives a single flow vector, and independent route choice at fixed probabilities ignores congestion feedback between travelers. Conditional stochastic user equilibrium (CSUE), in which each traveler of a finite population chooses a route by logit conditional on the routes of all others, contains this feedback but has been hard to compute with at scale. Under separable link costs and a common route-choice precision, the CSUE conditionals are those of a unique Gibbs distribution whose energy is a left-endpoint discrete congestion potential, and for continuous nondecreasing costs the law concentrates on deterministic logit SUE in the large-population limit. When the routes of each origin are the paths of an acyclic bush, retaining origin- or OD-level labels gives an unnormalized layered link-flow law with an explicit product-form multiplicity. On this representation we build a block Metropolis-Hastings sampler whose proposal is an integer version of Dial's loading under frozen costs. It leaves the CSUE law on the bush route sets invariant without enumerating paths or storing individual routes, and at stationarity the mean rejection rate of a block lies between one and two times the total variation distance between its proposal and exact conditional. In experiments, the sampler reproduces reference laws from exact enumeration and traveler-level Gibbs sampling, whereas independent route choice at frozen costs reproduces mean flows but overstates link-flow standard deviations by up to 48% and distorts correlations. On Sioux Falls, the block sampler attains 76.1 times the minimum effective sample size per CPU second of traveler-level Gibbs sampling, and on Chicago Sketch, with over one million travelers, its sampled covariance is used to assess traffic-count locations.
Kriste Krstovski, Sharyn O'Halloran, Kshitij Chandna
Wed 18 Nov
stat.MLcs.AIcs.CLcs.LG
pdf html
- reading document, topic model, government, discussion
- interest Language models as statistical objects 0.76
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.68
Abstract
Embedded topic models integrate word embeddings into the topic generation process, producing coherent topics and robust performance on large vocabularies. However, each document is represented by a single topic distribution, ignoring its internal structure. Many real world collections, including legal documents, company and government reports, and online discussion forums, exhibit hierarchical organization in which documents consist of multiple sections that often address distinct subtopics. We introduce the Hierarchical Embedded Topic Model (hETM), a structured topic model that couples the topic distribution of a document and those of its sections through a hierarchical prior in the embedding space. We derive an amortized variational inference procedure for hETM and evaluate it on structured collections from each of these domains. Experimental results show that hETM achieves the lowest or near-lowest held-out perplexity among flat and hierarchical topic model baselines, along with comparable or higher topic quality. It also performs strongly on a downstream clustering task, achieving the best clustering performance on most of the collections.
Joshua Joseph George, Hassan Safouhi
Tue 17 Nov
math.NAcs.NAmath.STstat.TH
pdf html
- reading derivatives, operator, inverse, sequences
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.75
- interest Kernels, operators and functional data 0.75
Abstract
Accurate computation of tail probabilities motivates efficient high-order implementations of the $G_n^{(1)}$ extrapolation transformation. We derive finite recurrence relations in the transformation order for Gaussian, gamma, and inverse Gaussian tail approximations. The Gaussian and inverse Gaussian numerator and denominator sequences satisfy four-term recurrences, whereas the gamma sequences satisfy a three-term recurrence. The derivation exploits the structure of the underlying differential operator and shows that the reduced numerator and denominator obey recurrences with identical coefficients. This construction avoids the repeated evaluation of higher-order derivatives and nested sums required by the explicit formulations. As a result, generating the approximation sequence through order $n$ requires $O(n)$ arithmetic operations, compared with $O(n^3)$ for the Gaussian and gamma distributions and $O(n^4)$ for the inverse Gaussian distribution. The recurrence algorithms also require only a fixed number of previously computed quantities at each order. Numerical results for the three distributions are in excellent agreement with values reported in the literature
Dmitry Ignatov
Thu 19 Nov
stat.MLcs.ITcs.LGcs.PFmath.ITmath.OC
pdf html
- reading theory, minimax regret, differential, prices
- interest Computational methods for heterogeneous agent and dynamic economic models 0.74
- your work near “Timing as an Action: Learning When to Observe and Act” 0.72
Abstract
Comparing AI models across resource budgets yields quality-resource frontiers relating error to latency, parameter count, training compute, and energy. Offset power laws summarize such frontiers, but their fitted parameters reveal neither the price-independent geometry nor the manner in which trade-offs extend across several resource channels. This paper addresses both gaps by developing a differential theory in which this structure is made explicit. Each frontier admits a price-independent plane-curve representation relating the effective number of active resource channels, $N_{\mathrm{eff}} = s^2/\kappa$ (squared slope over curvature), to the remaining log-error depth, $\Delta S = \ln(\varepsilon/\varepsilon_{\infty})$. This representation separates intrinsic shape from the scale and origin of the resource axis. Complete monotonicity and divisibility yield differential inequalities and frontier bounds. When each of $N$ resources contributes one active channel, the theory gives the parameter-free operating rule $N_{\mathrm{eff}} = 1$ in delivered units and derives minimax resource prices $\lambda^{\star}_i = \alpha_i/(N+1)$ with an exact worst-case regret of $\ln(N+1)$ nats. The results make it possible to interpret measured frontiers and compare resource costs without hand-chosen exchange rates. The theory is evaluated on 117 published frontiers and implemented in the open-source package nn-rank, which derives prices from measured quality and cost, ranks candidates, and reports supported choices and minimax regret: https://github.com/ABrain-One/nn-rank
Qunhui Guo, Jiongbai Liu, Md Noor E Alam
Fri 13 Nov
stat.APcs.LG
pdf html
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.74
- reading social, log, double, unemployment
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.67
Abstract
Drug poisoning mortality (DPM) remains a leading cause of death among working-age Americans, and its burden falls unevenly across communities. Existing county-level studies of social vulnerability and overdose have largely reported predictive or associational relationships, while the shape of the adjusted exposure-response relationship between overall social vulnerability and mortality remains unclear. We examined the CDC/ATSDR Social Vulnerability Index (SVI) and county drug poisoning mortality across three two-year windows spanning 2016-2021. The associational analysis compared linear regression, Random Forest, and XGBoost models of log mortality using period-specific SVI subfeatures and county covariates. The selected XGBoost models were interpreted with SHAP values. The double machine learning (DML) analysis estimated the exposure-response relationship between SVI and DPM and tested the linearity restriction. Socioeconomic vulnerability contributed positively to predicted mortality, with disability and unemployment among the strongest subfeatures, while racial and ethnic minority composition contributed negatively in all three periods. Summing the four theme contributions reduced the overall-index attribution by 34% to 36%, indicating substantial offsetting. Linearity was rejected in every window ($p < 10^{-26}$). The adjusted relationship increased through the middle of the SVI distribution and declined in the upper tail. This pattern remained stable both before and after the outbreak of the epidemic. Overall SVI therefore captured meaningful variation in overdose mortality but did not function as a uniformly increasing risk scale across its full range. Component-level interpretation and current overdose surveillance provide important context for interpreting the composite index.
Eduardo Caruso Barbosa Pacheco, Reynaldo Pereira Martins, Alexandre dos Santos Gualberto, Joao Vitor Souza Germano, Milton Miranda Neto, Francisco Louzada
Fri 13 Nov
stat.AP
pdf html
- reading appear, document, empirical, national
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.72
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.69
Abstract
Two recent studies on ship turnaround time at Brazilian ports report models with R-squared scores above 0.80 on data from the national waterway transportation agency. We audit their published datasets and reveal these results are inflated by structural data leakage. Specifically, 97 percent of berthing events in their random row-level split appear in both training and testing partitions. We demonstrate that a model's R-squared drops from 0.82 to 0.20 when the data is properly split at the berthing level. We further document additional methodological flaws, including survivorship bias, undocumented target truncation, and the retention of variables that algebraically reconstruct the target. To address these issues, we propose a reproducible pipeline that aggregates cargo to the berthing level, utilizes conditional imputation, and represents high-cardinality cargo variables through weighted averages of multilingual sentence embeddings. Evaluating six time targets and thirty seeds, our pipeline reveals that cargo embeddings yield statistically significant improvements on four operational targets, while administrative waiting times remain unaffected. Prior works missed this per-target heterogeneity by conflating genuine signal with leaked identity.
Zhijun Zhu, Haoyu Yang, Ruoyu Xiong, Bingjie Li, Zhigang Yao
Fri 13 Nov
math.STstat.TH
pdf html
- favorites near “Monte Carlo Geometry Processing” 0.80
- interest Kernels, operators and functional data 0.81
- reading procedures, manifold, expansions, hausdorff
Abstract
Local manifold-fitting procedures smooth noisy observations over neighborhoods of scale $h$, inducing a systematic normal bias of order $h^2+\sigma^2$ that persists at the population level. We develop geometric bias cancellation (GBC), a multiscale framework for higher-order manifold recovery under ambient Gaussian noise. For procedures admitting a normal-bias expansion $c(h,\sigma)b(x)+O(h^4)$, we construct cross-scale correspondence using estimated normal fibers and show that second-order normal accuracy suffices for fourth-order correspondence accuracy. Canceling the leading geometric bias yields $O(h^4)$ population Hausdorff accuracy while preserving an embedded manifold diffeomorphic to the underlying manifold. We establish the required expansions and finite-sample recovery guarantees for successive tangent-space projection and Gaussian density ridges. For STSP, geometric bias cancellation also achieves an asymptotically vanishing Hausdorff error ratio relative to oracle bandwidth-tuned single-scale fitting. For the local contraction estimator of \citet{yao2023manifold}, we identify an obstruction to scalar bias cancellation and develop a directional cancellation achieving $O(\sigma^4)$ population Hausdorff accuracy with uniform positive reach, as well as fourth-order finite-sample accuracy when $n\gtrsim\sigma^{-(d+6)}\log(1/\sigma)$. Simulations confirm the predicted improvement from second- to fourth-order population bias.
Yichen Lou, Mingyue Du, Yuxiang Wu, Jianguo Sun
Tue 17 Nov
stat.MEstat.CO
pdf html
- reading functional, consistent, practice, estimation heterogeneous
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.79
- favorites near “Survival Analysis via Ordinary Differential Equations” 0.86
Abstract
Cox or proportional hazards model is the most commonly used regression model for failure time data and has been extended in various ways. In this paper, we consider inference for a heterogeneous and functional Cox model with the focus on estimation of the latent heterogeneous or subgroup structure based on interval-censored failure time data. For the problem, we propose a novel sieve maximum likelihood estimation procedure with the uses of B-splines and Bernstein polynomials. For the implementation of the proposed method, an EM algorithm is developed and the number of subgroups can be determined in a data-driven manner. Also the resulting estimators are shown to be consistent and asymptotically normal, and an extensive simulation study shows that the proposed method works well in practice. Finally, we apply it to an Alzheimer's Disease study that motivated this investigation and provide some new insights about the roles of imaging and clinical factors in the progression of Alzheimer's disease.
Yichen Zhu, Guanhua Fang, Ming Zheng
Wed 18 Nov
stat.MEstat.APstat.ML
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.79
- interest Reinforcement learning, policy learning and off-policy evaluation 0.77
- reading higher, preferences, social, crossed
Abstract
Different paths need not imply different preferences, trajectories also differ in how randomly actions are chosen. Many multi-reward inverse reinforcement learning (IRL) methods assume common action randomness and can confound it with preference heterogeneity. We introduce TRAIL-IRL, which clusters unlabeled trajectories using rewards for preferences and temperatures for action randomness. Its crossed mixture pairs each reward with a shared set of temperature levels, allowing action regularity to vary within preference groups. With centered and normalized rewards in a known finite-state environment, we establish the identifiability conditions under which the distribution of sufficiently long trajectories determines rewards, temperatures, mixing weights, and class counts up to relabeling. Under the same conditions and reward constraints, common-temperature reward mixtures with the same component count cannot reproduce a crossed mixture with multiple temperatures. We develop a generalized expectation-maximization algorithm and select class counts using the Bayesian information criterion (BIC). Simulations show improved predictive fit, reward recovery, and trajectory classification over common-temperature mixtures with the same component count. For 140 bird-year trajectories from 79 white storks, BIC favors four reward classes and three temperature levels over common-temperature candidates. Younger birds have higher posterior membership probabilities for a southwestward migration class, whereas older birds have higher probabilities for a northern retention class. Within reward classes, higher temperatures are associated with more variable movement directions and positions farther ahead along the migration route, consistent with social information use and collective sensing.
Kuancheng Ye
Tue 17 Nov
stat.MEstat.ML
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.78
- interest Adaptive data collection and anytime-valid inference 0.82
- favorites near “Eluder Dimension and the Sample Complexity of Optimistic Exploration” 0.81
Abstract
We construct confidence intervals for linear functionals $\theta(F_m)=\int h\,dF_m$ of the empirical law $F_m$ of success probabilities $p_1,\dots,p_m$, from independent counts $c_i\sim\mathrm{Bin}(n,p_i)$, as for a language model run $n$ times per problem. Coverage is at least $1-\alpha$ for every $n$, $m$ and fixed $(p_i)$, simultaneously for all bounded $h$. The variance-adaptive (VA) region tests each candidate mixing law with the conditional variance of the tail counts it implies. This variance is linear in the law, so the region is a linear program, and cell envelopes keep it valid off-grid. Regions calibrated for iid sampling target a super-population law and are conservative for $\theta(F_m)$. With all $n$ tail counts, fixed $n$, continuous $h$ and a grid refined as $m\to\infty$, the identification part of VA's width is first-order worst-case optimal, as for any localization with the same index set. Under a rule fixed before the main comparator runs, VA was narrower than a two-stage plug-in band at every $n$ (median width ratio 0.60 at $n=128$), but not than the asymptotic, pointwise AMARI interval at $n=128$; on real rollout pools it was wider than both pointwise and Bonferroni-adjusted AMARI at $n\le128$. For $P(p>0)$ and pass@$k$ with $k\gg n$ the data support only lower bounds.
Rui Qiu, Yiwei Yao, Zhou Yu
Fri 13 Nov
math.PRmath.STstat.TH
pdf html
- interest Kernels, operators and functional data 0.80
- favorites near “Eigendecompositions of Transfer Operators in Reproducing Kernel Hilbert Spaces” 0.82
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.75
Abstract
Moving cut loci can change the limiting covariance of sample Huber means even when the cut locus of the population mean has probability zero. On a closed connected Riemannian manifold, we prove that the averaged logarithm field is $C^1$ near a point $m$ whenever the sampling law has a continuous density near its cut locus. The result holds for every continuous weight depending only on the observation, and the covariant derivative includes an explicit integral over the regular cut locus. When $m$ is a Huber mean for a fixed positive threshold, assume that the distance to $m$ has no atom at that threshold. The Hessian of the population risk then equals the expected pointwise Hessian minus a positive semidefinite cut locus correction. If $m$ is unique and this Hessian is positive definite, every measurable selection of global minimizers for i.i.d. samples satisfies a tangent-space central limit theorem. The cut loci of points in a ball of radius $\delta$ sweep out a region of volume $O(\delta)$, uniformly in the centre. These results establish the volume and integrability conditions previously assumed in a central limit theorem for intrinsic Fr\'echet means. At each fixed centre, the volume measure on this region, divided by $\delta$, converges weakly to an explicit measure on the regular cut locus. The Huber correction vanishes exactly when the region swept out about $m$ has probability $o(\delta)$.
Md Zulfiqur Haider, Asmaa Eldesoukey, Abhishek Halder, Yongxin Chen
Tue 17 Nov
math.OCcs.AIcs.LGcs.SYeess.SYstat.ML
pdf html
- reading control, aggregate, interacting, equation
- interest Computational methods for heterogeneous agent and dynamic economic models 0.74
- favorites near “Sequential Monte Carlo Steering of Large Language Models using Probabilistic Programs” 0.74
Abstract
We consider stochastic token steering through a deep stack of transformer layers as a mean-field control problem for an interacting particle system on the unit sphere, where interactions arise from the attention mechanism. In the mean-field limit, the token population evolves via a controlled integro-partial differential equation where the feedback control acts as a corrective vector field guiding the population toward an aggregate configuration of tokens in the representation space. We deduce the minimum effort control to accomplish this task over a fixed time horizon interpreted as normalized layer depth. We show that the resulting controlled dynamics for the token population can be understood as a mean-field Schr\"odinger bridge adapted to the sphere geometry. Building on this, we propose a novel generalization of the dynamic Sinkhorn recursion to compute the consistent token evolution. Numerical experiments confirm that the proposed approach achieves exact steering of the token population.
Davide Wiest
Fri 13 Nov
cs.LGstat.ML
pdf html
- favorites near “Fit without fear: remarkable mathematical phenomena of deep learning through the prism of interpolation” 0.81
- interest Kernels, operators and functional data 0.80
- your work near “Computational Methods for Economic Models with Function Valued States” 0.74
Abstract
Regularization-based continual learning reduces catastrophic forgetting by penalizing changes to parameters that mattered for earlier tasks. Elastic Weight Consolidation (EWC), a standard instance, weights its quadratic penalty by the empirical Fisher, usually justified as approximating loss curvature. We instead derive the penalty from what training on a new task should preserve. In this formulation, Continual Learning as an Optimization Constraint (CLOC), new training must keep chosen functionals of learned behavior unchanged, and the local approximation of this constraint is the stored penalty. We show that EWC's empirical Fisher is exactly the first-order form of this constraint when the preserved functional is each old sample's loss. The derivation exposes a flaw: EWC weights each sample by the scale of its loss gradient, so samples the model fits well barely contribute, although later training can still forget them. Improved EWC (IEWC) instead penalizes how far each old output moves off its loss level set, rather than how much its loss changes, which removes this weighting for any differentiable loss and output geometry. Across classification, segmentation, language, forecasting, and diffusion, IEWC variants match or improve on EWC and outperform EWC-DR, a classification-specific correction. In a held-out evaluation with validation-selected hyperparameters, removing this weighting alone raises ImageNet-R accuracy over EWC by 2.00 percentage points (95% CI 0.93-3.07). More broadly, CLOC turns the design of continual-learning penalties into a choice of what to preserve, from which a principled penalty follows by derivation.
Ying Chen, Zhangyang Wang, Yixuan He
Fri 13 Nov
cs.LGstat.ML
pdf html
- favorites near “Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss” 0.84
- interest Computational methods for heterogeneous agent and dynamic economic models 0.74
- reading pretraining, theory, lp, little
Abstract
Link prediction (LP) is a standard pretraining objective for graph neural networks (GNNs): hide edges, train an encoder-decoder to predict them, and reuse the node embeddings for downstream node tasks. Despite its empirical success, there is little quantitative theory explaining why an edge-level objective should produce embeddings that are linearly useful for node classification. We study this question in the finite-sample stochastic block model (SBM). For any bounded GNN encoder paired with a symmetric bilinear decoder, we prove a conditional reduction: small population held-out LP excess risk implies small community-classification error for a transductive linear classifier on the learned embeddings. The bound decomposes into an optimization residual, an architectural approximation residual, and explicit statistical terms of order $\tilde{O}(\sqrt{d/n})$ in bounded dense regimes. We further give a constructive few-label classifier: nearest-centroid classification under the positive semidefinite quadratic form $G = W Z^\top Z W$, whose label error is coupled to LP quality. The framework is architecture-agnostic through the approximation residual, and we instantiate this residual as $O(\log n / n)$ for a graph isomorphism network (GIN) with spectral positional encoding and a bilinear decoder. Synthetic SBM experiments validate the theorem chain from LP risk to score error to classification error, and show when the worst-case constants become conservative.
Timothy C. Pearce, David J. T. Smith, Alec Dobney, Alessia Freddo
Tue 17 Nov
physics.soc-phphysics.ao-phstat.AP
pdf html
- reading reports, did, persistence, community
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.71
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.71
Abstract
Ambient air quality monitors (AAQMs) measure exposure to emissions from waste facilities at few points; for most residents the only record is their own odour reports, used by regulators as evidence of impact but not typically as measurements of exposure. Calibrated public reports are accepted measurements in seismology, but whether community odour reports can serve as a sensor of exposure, and be forecast, is untested against independent monitors. Here we show they can, at a UK landfill, with SENTINEL, which combines the public weather forecast, run through a dispersion model, with the reports already received and no instrument at the site. Judged against the independent monitors, monthly report frequency was correlated with hydrogen sulphide (r = 0.92) and daily reports remained associated with it beyond weather; they fell fourteenfold after a change at the source while modelled transport did not; and, with no gas data as input, gave exposure in both monitored postcodes within a factor of three of that measured. SENTINEL forecast daily report counts with skill 0.64 one day and 0.50 four days ahead, exceeding damped persistence four days ahead; the one-day gain over persistence was unresolved. Community odour reports can act as a relative sensor complementing sparse monitoring; a monitor is needed for absolute scale.
Andrei Shelopugin, Alexander Sirotkin
Fri 13 Nov
cs.LGstat.AP
pdf html
- reading national, expectations, forecasting, matches
- favorites near “Recent Developments in Machine Learning Methods for Stochastic Control and Games” 0.76
- interest Language models as statistical objects 0.74
Abstract
Comparing soccer teams that rarely or never meet is a fundamental challenge in sports analytics: clubs play mostly within their national leagues, and national teams within their confederations. We propose a rating framework that places clubs and national teams worldwide on a common probabilistic scale. Building on a recent extension of Glicko with order effects and strength-dependent draw probabilities, we adapt the model to soccer with goal-scoring adjustments, a decomposition of home advantage, extra time, league-aware rating initialization and rules of promotion and relegation that do not inflate the levels of leagues; national teams start from the levels of their confederations. The models are evaluated on about 1.2 million club matches of 252 leagues from 166 countries (2000-2026) and 12.6 thousand national team matches (1998-2026). On held-out seasons, the modified Glicko model is clearly more accurate than Elo and as accurate as score-matrix models built on the goal expectations of a gradient boosting model, including the Dixon-Coles and Markov chain models, while it is substantially more accurate in international club matches. Averaging the predictions of the rating model and the Markov chain model is more accurate than either model alone. For national teams, the model is significantly more accurate than all baselines, especially in the final tournaments, where teams of different confederations meet. The resulting ratings are interpretable and can be used for player evaluation, transfer analysis and match forecasting. The match data, the ratings and the code are publicly available at https://github.com/andreyshelopugin/SportsRatings
Shijie Wang, Haichao Gui, Rui Zhong
Fri 13 Nov
stat.MEcs.RO
pdf html
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.77
- interest Bayesian computation and structural estimation 0.80
- favorites near “Inference with Hamiltonian Sequential Monte Carlo Simulators” 0.76
Abstract
This paper proposes a novel class of Bayesian Conjugate Distributions (BCDs) on extended special Euclidean groups SE_n(3) for global motion uncertainty representation and estimation. BCDs are constructed by restricting matrix Gaussian distributions on ambient Euclidean space to SE_n(3), avoiding singularity of local coordinates. The unique feature of BCDs is their conjugacy to a broad class of sampling distributions arising in motion estimation, enabling exact closed-form Bayesian inference. Optimal estimation with respect to BCDs is further investigated for implementing motion estimation. Two means are extracted as random state estimates, which maximize probability and minimize mean squared error, respectively. Moreover, within the BCD class, moment matching is shown to minimize the Kullback-Leibler divergence, and maximum likelihood estimation is also derived, which provide analytical methods to approximate underlying distributions with the optimality guarantee. As an application, a Bayesian estimator for pose and velocity estimation of an Unmanned Aerial Vehicle (UAV) is developed within an assumed density filtering architecture. The propagation and estimate extraction leverage the optimal estimation related to BCDs, and the landmark measurements are exactly fused with the BCD prior in closed form due to the conjugacy. Numerical simulations of a UAV flight indicate the advantages of BCD-based estimators over two classic Kalman filters on Lie groups with various initial uncertainties and measurement noise.
Joshua Foster, Fredrik Odegaard
Tue 17 Nov
cs.LGecon.EMstat.ML
pdf html
- reading auction, language model, prices, online
- interest Language models as statistical objects 0.76
- favorites near “Recent Developments in Machine Learning Methods for Stochastic Control and Games” 0.73
Abstract
Online auction houses must set reserve prices for highly differentiated products whose unstructured descriptions contain valuable demand information. Existing approaches generally require managers to specify product attributes in advance. We ask whether a language model can instead learn from listing descriptions and historical bids to set item-specific reserve prices before bidding begins. We fine-tune a pretrained text embedding language model using the structural likelihood of auction bids as its training objective. This directs representation learning toward the valuation distribution that determines the reserve decision. We show that consecutive upper bids identify the tail information needed to set an optimal reserve without observing or modeling the number of potential bidders. A split-sample estimator learns the representation on one half of the data and conducts structural estimation and inference conditional on that representation on the other. We apply the method to approximately 65,800 online used-vehicle auctions. With seller values calibrated to match the sale rate among listings, the model implies a counterfactual net increase of $165.8 million, representing a 14.6% increase above the incumbent reserve policy. The framework translates a listing description and a seller's value into an item-specific reserve recommendation without constructing a product taxonomy or requiring a bid history for the item being priced. It provides an approach to adapting language models for operational decisions in marketplaces where products are heterogeneous and demand-relevant information is recorded in text.
Christopher Marouani, Grigory Bartosh, Christian A. Naesseth
Mon 16 Nov
stat.MLcs.LG
pdf html
- favorites near “Spectral State Space Models” 0.77
- interest Language models as statistical objects 0.80
- your work near “Valid Inference with Imperfect Synthetic Data” 0.73
Abstract
Contrastive representation learning methods operate by pulling different views of the same data sample together while pushing them away from views of different samples. A fundamental challenge is the estimation of the repulsive force, which typically requires large batch sizes, can exhibit high variance, and is susceptible to the curse of dimensionality. We introduce ScoreCLR, a contrastive representation learning method that offers a novel way to estimate this repulsion term. We consider the framework of representation learning by maximizing the mutual information between two latent views of a sample. By deriving the gradient of the Shannon entropy with respect to the encoder parameters, we obtain an equivalent objective that involves the score of the latent representation. We then use denoising score matching techniques to estimate this score with a lightweight network that is trained jointly with the encoder. We further propose regularization techniques using different noise levels and a Gaussian prior on the latent representation. ScoreCLR replaces the batch of explicit negative samples in contrastive methods with a learned repulsion whose cost is linear in the batch size rather than quadratic. On CIFAR-10 and ImageNet-100 it does so with minimal loss of linear-probe accuracy compared to SimCLR under a shared recipe, while producing projector outputs of higher effective rank, closer to the Gaussian prior.
Kun Huang, Xiyu Peng, Huiyan Sang
Mon 16 Nov
stat.ME
pdf html
- reading large, resulting model, graphs, consistent
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.77
- favorites near “Random Walks on Simplicial Complexes and Harmonics” 0.79
Abstract
Bayesian graph-informed random partition models are widely used to model heterogeneity in structured data. A prominent approach assigns a prior directly on the space of contiguous partitions of a pre-determined graph G; however, such models are not data-adaptive and can be computationally infeasible for large graphs. To address these limitations, we propose a Bayesian adaptive quotient graph framework that induces partitions through a two-level hierarchical prior: a prior is first imposed on quotient graphs of G, followed by a conditional prior on a reduced contiguous partition space induced by each quotient graph. We develop high-level assumptions that guide the construction of the two-level prior such that the reduced space retains desirable partitions. We further establish high-level assumptions for posterior concentration on desirable quotient graphs and partitions. The framework can be readily embedded into many existing Bayesian hierarchical models with latent cluster structures as a contiguous random partition prior. As a specific example, we apply the proposed framework to the spatial random partition model proposed in Huang and Sang (2026). The resulting model adapts to the unknown number of clusters, achieves a faster clustering-consistency contraction rate, and yields substantially lower partition error and computation time in simulations. Finally, we apply the resulting model to a large spatial transcriptomics dataset to demonstrate its practical utility.
Luca Anzilli, Fabrizio Durante
Fri 13 Nov
math.STstat.MEstat.TH
pdf html
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.77
- interest Semiparametric inference and debiased machine learning 0.82
- reading rectangles, describes, practice, parametrized
Abstract
The average rectangular volume ($\ARV$) of a bivariate aggregation function is the average of its volumes over all axis-parallel rectangles of the unit square; for a copula, it is the expected copula volume of the rectangle spanned by two independent uniform points. We determine the exact region of the pairs $(\ARV(C),\rho(C))$, where $C$ ranges over all bivariate copulas and $\rho$ is Spearman's rho. The region is compact, convex and symmetric, and its upper boundary consists of three analytically parametrized arcs. A previously proposed family of shuffles of Min generates one arc, but only on part of its range; the other two arcs are generated by singular copulas whose supports contain hyperbolic branches. The boundary is continuously differentiable away from its left endpoint $\ARV=\frac{11}{120}$, where it has a vertical tangent. The proof uses a folding transformation and explicit Kantorovich potentials for a family of optimal transport problems. Finally, we show that the arachnitude, a rank correlation used in actuarial practice, is an affine transformation of $\ARV$, so that the result also describes the joint region of Spearman's rho and arachnitude.
Innocent Abaa, Umar Islambekov
Fri 13 Nov
stat.MLcs.LGstat.AP
pdf html
- favorites near “Hierarchies of Relaxations for Online Prediction Problems with Evolving Constraints” 0.81
- reading data analysis, filter, functions, higher
- interest Online learning and statistical learning theory 0.78
Abstract
We propose a nested ensemble learning framework that utilizes Topological Data Analysis (TDA) to extract and integrate topological features from graph data, with the goal of improving performance on classification and regression tasks. Our approach computes persistence diagrams (PDs) using lower-star filtrations induced by three filter functions: closeness, betweenness, and degree2 centrality. To overcome the limitation of relying on a single filter, these PDs are integrated through a data-driven, three-level architecture. At Level-0, diverse base models are independently trained on the topological features extracted for each filter function. At Level-1, a meta-learner combines the predictions of these base models for each filter to form filter-specific ensembles. Finally, at Level-2, a meta-learner integrates the outputs of these filter-specific ensembles to produce the final prediction. We evaluate our method on both simulated and real-world graph datasets. Experimental results demonstrate that our framework consistently outperforms base models and standard stacking methods, achieving higher classification accuracy and lower regression error. It also surpasses existing state-of-the-art approaches, ranking among the top three models across all benchmarks.
Xingtu Liu
Tue 17 Nov
stat.MLcs.LGmath.STstat.TH
pdf html
- interest Computational methods for heterogeneous agent and dynamic economic models 0.77
- your work near “Timing as an Action: Learning When to Observe and Act” 0.76
- favorites near “Finding Regularized Competitive Equilibria of Heterogeneous Agent Macroeconomic Models via Reinforcement Learning” 0.74
Abstract
Gaussian approximation provides a foundation for online statistical inference in reinforcement learning. We establish sharp non-asymptotic Gaussian approximation rates in Wasserstein distance for asynchronous Q-learning with Polyak-Ruppert averaging. We prove an upper bound of $\widetilde{O}(n^{-1/4})$, where $n$ is the number of iterations, together with a matching lower bound up to logarithmic factors. Q-learning is more challenging to analyze than linear stochastic approximation due to the nonlinearity of the Bellman operator. Asynchronous updates introduce further challenges due to Markovian noise. Our analysis separates the errors from Bellman nonlinearity, Markovian noise, and the finite-sample covariance discrepancy. In particular, we show that Markovian noise induces an unavoidable bias even when the Bellman operator is linear. The trade-off between this bias and the covariance discrepancy determines the convergence rate. We further extend our analysis to the synchronous setting and establish a sharp rate of $\widetilde{O}(n^{-1/2})$.
Soukaina Douissi
Thu 19 Nov
math.STmath.PRstat.TH
pdf html
- favorites near “Optimal uniform convergence rates and asymptotic normality for series estimators under weak dependence and weak conditions” 0.86
- interest Kernels, operators and functional data 0.78
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.74
Abstract
We study the estimation of the drift parameters $(a,b)$ of the fractional Vasicek process $dX_t=a(b-X_t)\,dt+dB^H_t$, driven by a fractional Brownian motion of Hurst index $H\in(0,1)$. The estimators are of moment type, built from the empirical mean and variance of the path, and are analysed in two regimes: continuous observation on $[0,T]$, and high-frequency observation at times $t_i=i\Delta_n$ with $\Delta_n\to0$ and $T_n=n\Delta_n\to\infty$. Using the Malliavin--Stein method on the second Wiener chaos, we prove consistency of both estimators for every $H$, and quantitative central limit theorems in the Wasserstein distance: for $H<\frac34$ for the estimator of $a$, with explicit rates depending on $H$, and for every $H\in(0,1)$ for the estimator of $b$, at the rate $T^{-H}$. In the discrete regime the bounds contain an additional discretization term of order $\Delta_n^{\min\{H+1/2,\,1\}}$, and for $H<\frac12$ the central limit theorem for the estimator of $b$ requires in addition $T_n^{1-2H}\Delta_n^{1+2H}\to0$. A short Monte Carlo study illustrates the results.
Lisa Chang, Yijun Wang, Shanshan Ding
Mon 16 Nov
stat.APstat.ME
pdf html
- reading data analysis, standard errors, makes, tend
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.79
- interest Semiparametric inference and debiased machine learning 0.81
Abstract
This article studies the envelope model for enhanced gene expression analysis using GTEx data. The study compares this advanced statistical approach with standard multivariate linear regression, demonstrating improved capacity for identifying significant genes and enhanced estimation efficiency. By partitioning response variation into material and immaterial components, the envelope model reduces estimation noise, leading to stronger statistical signals and smaller standard errors. These findings, as demonstrated in our tissue-specific and gene-group anal- yses, show that the envelope model can pick up on important biological signals that traditional methods tend to miss. This makes it a useful tool for genomic studies and GTEx data analysis.
Haoxuan Wang, Jianliang Ye, Sen Na
Thu 19 Nov
stat.MLcs.LGmath.OCstat.CO
pdf html
- interest Computational methods for heterogeneous agent and dynamic economic models 0.78
- your work near “Efficient Online Estimation of Causal Effects by Deciding What to Observe” 0.79
- favorites near “Recent Developments in Machine Learning Methods for Stochastic Control and Games” 0.74
Abstract
We study statistical inference for online sketched Newton methods, in which each Newton system is approximately solved by an inner randomized sketching solver. Existing work reveals a computational--statistical trade-off: the sketching solver (and its accelerated variants) reduces the per-step cost of Newton methods from $O(d^3)$ to $O(d^2)$, but the randomness introduced by sketching also inflates the limiting covariance of both the last and the averaged Newton iterates. In this work, we address this trade-off via gradient averaging. We show that simply replacing the stochastic gradient with a moving average at each step can mitigate, and in some cases eliminate, the trade-off in Newton sketch. Importantly, gradient averaging requires no additional oracle evaluations and leaves the contraction bound of the inner sketching solver unchanged. In particular, we prove that the last sketched Newton iterate with gradient averaging attains a limiting covariance no larger than that without gradient averaging, while the averaged iterate even attains the minimax optimal limiting covariance, matching that of the exact Newton method. To our knowledge, this is the first analysis showing that randomness introduced to improve computational efficiency need not compromise statistical efficiency as long as proper gradient estimates are used. The key analytical challenge is that gradient averaging destroys the crucial martingale difference structure underlying central limit theorem arguments. We overcome this challenge by analyzing the coupled two-timescale recursions governing the iterate and the gradient estimator. Numerical experiments on linear and logistic regression support our theoretical covariance comparisons.
Henry Bayly, Yorghos Tripodis, Haolin Li, Prasad Patil
Thu 19 Nov
stat.ME
pdf html
- reading did, available, models developed, large
- interest Language models as statistical objects 0.75
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.73
Abstract
Survival prediction models developed in multi-site studies may transport poorly to a new site because of between-site heterogeneity in covariate effects and baseline hazards. Pooled models may fail to capture site-specific risk patterns, while site-specific models can be unstable in small cohorts. We propose a Multi-Site Super Learner (MSSL), a domain-adaptive ensemble for survival prediction that combines diverse pooled and site-specific base learners. A meta-learner optimizes the ensemble weights using source data alone or, if available, a small target-site calibration sample. In simulations, when even a small target-site sample was available, MSSL substantially reduced prediction error relative to a pooled super learner under moderate to large between-site heterogeneity and generally matched or improved upon models fit to the target-site data alone. It did so by downweighting dissimilar or noisy models and upweighting models from structurally similar sites. Without target-site data, MSSL did not improve upon the pooled super learner. In an application to predicting dementia progression using the National Alzheimer's Coordinating Center (NACC) Uniform Data Set across 10 sites, MSSL matched or outperformed standard super learning using pooled data in 9 of 10 sites. MSSL offers a practical framework for adapting multi-site survival prediction models to a new site when a small calibration sample is available.
Luiza L. S. Piva, Matheus V. Cerqueira, Francisco A. Rodrigues
Tue 17 Nov
cs.SIstat.CO
pdf html
- reading structural, consistent, large, seen
- your work near “Valid Inference with Imperfect Synthetic Data” 0.76
- interest Language models as statistical objects 0.76
Abstract
Network sampling is essential for the study of real world systems, since access to the topological structure and dynamical characteristics of a complex system is usually infeasible, and can also prove computationally intractable for large and dense networks. However, the field still lacks systematic methodologies that allow a procedural analysis of sampling plans for connected data. Hence, this study aims to compare network sampling methods through the usage of multivariate statistical analysis and machine learning techniques that are anchored in network metrics' estimation of synthetic and real-world networks. It was demonstrated that no single sampling strategy dominates both cases, although Forest fire and Frontier sampling have presented the most consistent performance across all topologies investigated. The impact of each structural property of these networks for principal component analysis and classification results were also discussed, with consistent pattern groups seen for both data types. Transfer learning aimed at real-world network identification has also demonstrated that realistic structural connections can be correlated to well-known network models through the usage of sampled subgraphs' properties for model training, providing an avenue to real-world network characterization through sampling methodologies.
Paul Wilsens, Katrien Antonio, Gerda Claeskens, Koos Gubbels
Fri 13 Nov
stat.APcs.LG
pdf html
- favorites near “Eight Centuries of Global Real Interest Rates, R-G, and the 'Suprasecular' Decline, 1311-2018” 0.71
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.76
- interest Macroeconometrics and time series 0.77
Abstract
We study the occurrence and reporting dynamics of weather-related claims in private home insurance, combining detailed portfolio and claim information with publicly available high-resolution weather data. Existing approaches typically start from an identified severe weather event, such as a hailstorm, and model the dynamics of the claims caused by that event. In practice, however, claims arise under moderate as well as severe conditions, and the claims handling process runs continuously. An insurer therefore requires insight into the occurrence and reporting of claims on every calendar day. Modelling all calendar days jointly poses two challenges: recently occurred claims are only partially reported, and days with severe weather are rare while accounting for a large share of the claims. We propose a multi-regime EM-XGBoost framework that addresses both: an expectation-maximisation setup accounts for the incurred-but-not-reported (IBNR) claims, while a data-driven regime structure lets the occurrence and reporting dynamics differ across weather conditions. In a case study on a Dutch home insurance portfolio, we compare single- and two-regime specifications for two spatial resolutions: the country as a whole, and its individual regions. We find that the regional two-regime model outperforms the regional single-regime model as well as both national models, indicating that the benefit of a multi-regime setting depends on a sufficiently granular spatial resolution. We also show that daily covariates constructed from hourly weather observations, capturing the duration and the intra-day timing of severe weather, improve both the occurrence and the reporting model.
Ivano Lodato, Simran Hiranandani, Aditya Varna Iyer
Mon 16 Nov
stat.MEstat.CO
pdf html
- favorites near “Simplifying debiased inference via automatic differentiation and probabilistic programming” 0.84
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.75
- reading written, make, formula, changes
Abstract
Identification formulas returned by the ID algorithm often contain summations that can be removed. The procedure proposed in arxiv:1806.07082 (TK) removes them by combining factors of the formula into joint distributions, through the chain rule and through the insertion of factors for missing variables. We make two changes to this procedure: only the factors that involve the summed variable are combined, and every factor is written with a minimal conditioning set. These two changes determine the rest of the procedure, in which each search over a power set is replaced by a single test and the variables to insert are fixed in advance, and they remove summations that the TK procedure leaves in place. The resulting procedure is sound and, for minimal input terms, complete for each individual elimination with respect to joins and insertions.
Peiyao Liu, Jie Guo, Zhijun Wang, Xueru Zhang, Nan Chen
Wed 18 Nov
stat.MEcs.ITmath.IT
pdf html
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.78
- interest Adaptive data collection and anytime-valid inference 0.81
- reading functional, functional data, basis, sequential
Abstract
Quickest change detection (QCD) is studied for partially observed functional data streams. At each sampling epoch, a function is realized over a continuous domain, but its values can be observed at only a fixed number of locations. At an unknown change point, the functional mean undergoes a shift that is assumed to admit a sparse representation in a localized basis. We propose Functional Adaptive Sensing and Testing (FAST), which jointly estimates the unknown post-change parameter, selects sensing locations, and performs sequential detection. A minimax lower bound on the worst-case detection delay is derived under an average-run-length constraint. A non-asymptotic estimation-error bound is established for the policy-dependent observations induced by adaptive sensing. Under the stated regularity conditions, FAST is first-order asymptotically optimal when sensing subsets are optimized exactly. Simulations and a wildfire case study demonstrate the performance of the proposed procedure.
Liuyuan Jiang, Prasanna Sattigeri, Lisha Chen
Mon 16 Nov
cs.LGmath.OCstat.ML
pdf html
- favorites near “Monte Carlo Geometry Processing” 0.79
- reading fewer, solutions, generally, nonlinear
- interest Computational methods for heterogeneous agent and dynamic economic models 0.73
Abstract
Constructing a finite Pareto representation requires allocating a limited number of solutions across the Pareto front (PF). A desired preference distribution specifies this allocation by assigning more solutions to some PF regions and fewer to others. Scalarization generates Pareto solutions by varying objective weights and solving single-objective problems. However, a distribution of scalarization weights generally does not translate into the same distribution of Pareto solutions along the PF, because the weight-to-solution map is nonlinear and unknown in advance.To account for this distortion, we introduce SURFACE (Scalarization Updates for Redistributing Finite Approximations with Chart Estimation}), which adaptively updates scalarization weights to match a desired PF distribution. SURFACE estimates how each region of the weight grid corresponds to an area of the PF from the current scalarized solutions, then redistributes the weights accordingly. The method supports parallel, warm-started, inexact optimization with standard single-objective solvers. Theoretically, we establish that, under the stated regularity conditions, SURFACE achieves linear convergence of the preference distribution mismatch to a finite-resolution error floor of $\mathcal{O}(N^{-2/(M-1)})$, where $N$ is the number of adaptive weights and $M$ is the number of objectives. Empirically, experiments across multiple benchmarks demonstrate improved target-mass balance and PF coverage compared with established baselines.
Olatunji Oni, Sanhita Sengupta, Marc De Benedetti, Nathanial Eddy
Tue 17 Nov
cs.LGq-bio.QMstat.AP
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.78
- favorites near “Bayesian Semiparametric Model for Sequential Treatment Decisions with Informative Timing” 0.82
- reading large, burden, treatment, llms
Abstract
Large language models (LLMs) have enabled the ability to scale clinical adjudication tasks traditionally dependent on exhaustive manual review. Line-of-therapy (LoT) adjudication, a major bottleneck in oncology cohort construction, is a prime example. Line boundaries must be inferred from longitudinal treatment histories according to nuanced, study-specific definitions, at every regimen transition for every patient. LLMs can apply this adjudication at scale, but their reliability varies across patients. Treatment histories differ in complexity, and ambiguous or atypical patterns depend on expert clinical judgment. Reliable automation thus requires verified support for model assignments and the ability to identify cases that warrant manual review. Here, we present evidence-calibrated selection (ECS), a human-in-the-loop framework for scaling LLM-assisted LoT adjudication. ECS derives confidence from observed clinician adjudications, prioritizes patients with the least-supported assignments for follow-up review, and returns completed adjudications to the model as precedent. As the evidence base grows, ECS continuously calibrates confidence to estimate residual error, thus enabling more efficient use of clinician time. In simulations on 136 clinician adjudicated patients with multiple myeloma, ECS achieved a residual error of 5% after review of 44.6% of patients, compared to 62.2% when review was directed by LLM self-reported confidence and 81.9% under random sampling. These results support a scalable approach to study-specific LoT adjudication, using clinician evidence to inform automated predictions, minimize error, and reduce review burden.
Nannan Zhou, Guanmeng Xian, Ning Yang
Tue 17 Nov
stat.MLcs.AIcs.LGphysics.flu-dyn
pdf html
- favorites near “Neural Operator: Learning Maps Between Function Spaces With Applications to PDEs” 0.80
- reading operators, state, manifolds, neural operators
- interest Computational methods for heterogeneous agent and dynamic economic models 0.74
Abstract
Autoregressive neural operators approximate time-dependent PDE solvers by repeatedly composing a learned transition, although the resulting states need not remain feasible under the discrete constraints of the numerical problem. We study this issue for incompressible flow, where each velocity state must satisfy solver-level incompressibility together with prescribed boundary values. We formulate autoregressive operator learning as a map between solver-defined affine feasible sets and introduce PCM-NO, a projected-residual realization of this feasible-transition principle. The neural backbone predicts a candidate state increment, a fixed tangent projection extracts its admissible component, and a metric retraction maps the advanced state to the prescribed feasible set. With exact affine constraint solves, the resulting transition is closed on the feasible set, while an affine identity separates admissible-update parameterization from state-level closure. With approximate solves, tangent-only residuals admit a cumulative upper bound, whereas per-step retraction resets feasibility to the residual achieved by the current solve. Experiments on periodic Navier--Stokes flow and lid-driven cavity flow demonstrate the superiority of PCM-NO in preserving solver-level constraints over long autoregressive rollouts and under parameter, boundary-value, and resolution shifts, while achieving lower rollout errors and better flow-specific diagnostics across both in-distribution and stress settings.
Haisheng Fu, Gexin Huang, Haikai Zhao, Jie Liang, Zhenman Fang, Jingning Han, et al.
Thu 19 Nov
cs.CVstat.ME
pdf html
- reading questions, question, generic, higher
- interest Language models as statistical objects 0.77
- favorites near “Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss” 0.76
Abstract
Attention maps cheaply show where a vision-language model (VLM) looks, but attention sinks make them unreliable for localization. Contrasting the question with a generic prompt suppresses sinks, yet only some layers localize the queried object. Existing methods find these layers with box annotations or fixed rules. We show that attention itself can identify them. A question changes how much a head attends to the image and where this attention goes. The two effects separate exactly, and the second, measured in bits on unlabeled questions, indicates which layers localize. WhereBits turns these where-to-look bits into a layer prior and combines it with per-head contrastive scores. Its three design constants were fixed once on labeled COCO data, and a new model or resolution needs only unlabeled questions. Across eight VLMs, WhereBits matches the pointing accuracy of box-tuned layers on COCO and RefCOCO and ranks target tokens better. This advantage holds on a held-out model family and at higher resolutions. Its COCO pointing accuracy is 0.72, against 0.22 for mean attention. In training-free zoom-in, its crops raise V* accuracy by 6.0 points under forced-choice evaluation.
Zhouxing Luo, Lei Wang, Xin Liu
Thu 19 Nov
math.OCcs.LGmath.STstat.MEstat.MLstat.TH
pdf html
- interest Kernels, operators and functional data 0.81
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.76
- favorites near “Variational Bayes under Model Misspecification” 0.84
Abstract
Principal component analysis (PCA) is a fundamental and ubiquitous technique for dimension reduction and representation learning, yet its empirical formulation remains vulnerable to distributional uncertainty. While distributionally robust optimization (DRO) offers a natural remedy, existing order-2 Wasserstein models of PCA either preserve the nominal PCA subspace or lack an exact reformulation. This paper demonstrates how the infinity-Wasserstein distance can reconcile these two desirable properties. For the resulting DRO model of PCA, we derive an exact finite-dimensional reformulation and explicitly identify a worst-case distribution attaining the inner maximum. Crucially, the reformulated objective retains samplewise residual information beyond the empirical covariance matrix, revealing a mechanism through which distributional robustness can reshape the learned principal subspace. Under the mild assumption that the true distribution has a finite fourth moment, we further establish a uniform finite-sample performance guarantee for the proposed model. An iteratively reweighted majorization-minimization algorithm is also developed to solve the resulting nonconvex and nonsmooth optimization problem, with a rigorous analysis of its convergence properties. Numerical experiments substantiate the computational efficiency of the proposed algorithm and the practical effectiveness of the infinity-Wasserstein model in downstream applications.
Gabriel Pena, Veronica Moreno, Nestor Barraza
Wed 18 Nov
cs.SEstat.AP
pdf html
- reading software, processes, theory, expectations
- interest Bayesian computation and structural estimation 0.77
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.69
Abstract
A review of several models for the software failure detection stochastic process is per- formed. We focus on how the Mean Time Between Failures is obtained from those models and present a comparative study. The models considered involve some of the well known based on non- homogeneous Poisson processes and a recently proposed model based on contagion. Our main target is to evaluate the models performance when they are applied to modern software projects developed under Agile methodologies. Recently developed Software Engineering practices (TDD, Scrum, etc.), propose an iterative procedure be- tween development, testing and debugging, introducing this way, particular challenges to Software Reliability modeling. As opposed to the non-homogeneous Poisson processes, the failure rate of the contagion model depends not just on time but also on the number of previously detected failures. All the considered models are particular cases of non- homogeneous Birth Processes, a general fundamental theory on how to obtain time to failure and time between failures expectations is also shown.
I. Elizabeth Kumar, Sherri Rose
Fri 13 Nov
stat.APcs.CYstat.ME
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.80
- interest Language models as statistical objects 0.73
- reading empirical, supplement, closely related, implicit
Abstract
Synthetic data augmentation has been proposed as a way to artificially supplement clinical datasets to address fairness concerns in downstream classification algorithms. This approach is closely related to upweighting or upsampling methods employed for the problem of class imbalance, which have shown mixed empirical results in practice. Synthetic data augmentation also runs the risk of overfitting to the noise present in subgroups with relatively small sample sizes, raising both generalizability and transportability concerns. We present an empirical evaluation of synthetic data augmentation for fairness in tabular data using simulation studies as well as a real-world application of in-hospital mortality classification for heart failure patients in two distinct data samples. We implement upsampling and upweighting methods as benchmarks for the implicit fairness objective of synthetic data augmentation. Our results suggest that the fairness impacts of data augmentation are sensitive to the design of the downstream algorithm as well as the deployment context and its similarity to the training data. It is crucial that studies making generalizability and transportability claims should provide empirical support for these assertions.
Rachel Li, Chris Hays, Aaditya Ramdas, Manish Raghavan
Fri 13 Nov
cs.AIcs.CYmath.STstat.TH
pdf html
- reading failure modes, stop, good, rise
- your work near “Timing as an Action: Learning When to Observe and Act” 0.75
- interest Adaptive data collection and anytime-valid inference 0.78
Abstract
In pre-deployment testing of AI systems such as AI agents, a developer can often deal with known failure modes (e.g., leaking API keys, deleting important databases, or installing a malicious package) by implementing fixes or making contingency plans for when the same issue might occur again. What is more challenging are the unknown unknowns: the unseen failure modes that haven't surfaced during testing. The decision to stop testing and deploy therefore faces an inherent tradeoff: continue testing to observe more behaviors, at the cost of time and resources, or stop testing and deploy, at the risk of failure modes remaining undiscovered. How then does the developer decide when they have tested enough? To answer this, we develop an algorithm that certifies when the probability of observing an unseen failure mode is sufficiently small. The algorithm constructs an anytime-valid upper confidence sequence on this quantity, so the developer can repeatedly test a system, inspect the bound, and decide to stop when it falls below a chosen threshold without invalidating the statistical guarantee. Our problem can be interpreted as a sequential variant of the missing-mass estimation problem which gave rise to the classic Good-Turing estimator (Good, 1953). On SWE-Chat, a dataset of real-world coding agent traces (Baumann et al., 2026), we implement our procedure and show that it achieves tighter upper bounds than a simple multiple-testing corrected Good-Turing baseline, and even dominates a fixed-horizon (non-multiple-testing-corrected) Good-Turing upper confidence bound, which is not anytime valid. Additionally, to better match realistic evaluation pipelines, we extend our method to handle sampling from a different task distribution and imperfect automated failure labeling.
Yian Huang, Jian Huang
Fri 13 Nov
cs.LGstat.ME
pdf html
- interest Computational methods for heterogeneous agent and dynamic economic models 0.77
- favorites near “Learning Scalable Deep Kernels with Recurrent Structure” 0.80
- your work near “Automated Solution of Heterogeneous Agent Models” 0.77
Abstract
Conditional generative modeling with limited target data can benefit from representations learned in a related source population. We develop a transfer learning framework for conditional flow matching that accommodates distinct source and target response distributions and dimensions. Two procedures exploit a representation learned from source data: TFM-Freeze estimates a target velocity head with the representation fixed, whereas TFM-Adapt jointly learns the head and a representation adaptation under a functional anchoring penalty. For explicit neural network classes, we establish bounds on conditional squared Wasserstein risk that account for source estimation, target fitting, representation mismatch, and numerical integration. When the source representation admits a smooth correction, adaptation replaces the frozen mismatch contribution by the cost of learning the correction and a controllable regularization bias. An information lower bound identifies regimes in which adaptation attains a strictly faster conditional squared Wasserstein risk rate than any generator restricted to the frozen context information. We also develop independent validation procedure for selecting among fitted conditional generators. We show that the selection procedure satisfies an oracle inequality, and the switching rule guarantees, with high probability, no increase in conditional energy risk relative to a target-only generator. Simulations and real data applications illustrate the gains from our proposed methods and the effectiveness of target validation in guarding against negative transfer.
Debsurya De, Dmitriy Kunisky
Fri 13 Nov
math.STcs.CCmath.PRstat.TH
pdf html
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.76
- interest Computational methods for heterogeneous agent and dynamic economic models 0.75
- favorites near “Introduction to regularity structures” 0.80
Abstract
We study high-dimensional central limit theorems (CLTs) as hypothesis testing problems, comparing the number of summands needed for statistical and computational indistinguishability of the normalized sum from the limiting Gaussian distribution as the dimension and number of summands grow together. For the off-diagonal entries of sums of i.i.d. rank-one tensors formed from standard Gaussian vectors (equivalently, the problem of distinguishing Wigner from Wishart tensors), we prove total variation bounds sharpening previous estimates of Mikulincer (2020) and give matching polynomial-time tests. This shows that the statistical and computational thresholds coincide for this CLT, contrary to a proposal by Kunisky, Moore, and Wein (2024). On the other hand, we also show that, for tensors of order at least three, sufficient Gaussian dilution of the summands by adding independent Gaussian noise and rescaling gives a CLT model with a statistical-to-computational gap (in the above setting and ones with more general rank-one summand distributions). To the best of our knowledge, such a phenomenon has not been observed before in any high-dimensional CLT. In particular, in the relevant parameter regimes, we show that a brute-force search achieves strong detection, while polynomials of degree up to $n^{\Omega(1)}$ cannot achieve strong separation. The latter lower bounds control low-degree $\chi^2$-divergences by a novel method based on cumulant generating functions, and we further illustrate the use of these tools by applying them to masked tensor CLT models where only a prescribed subset of tensor entries is revealed.
Avi Vajpeyi, Giorgio Mentasti, Ollie Burke, Renate Meyer, Patricio Maturana-Russel, Quentin Baghi
Tue 17 Nov
stat.MEastro-ph.IM
pdf html
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.77
- interest Macroeconometrics and time series 0.79
- reading spectrum, tensor-product, instrumental, likelihood
Abstract
We model the evolutionary spectrum of a slowly time-varying process by a linear combination of products of time- and frequency-dependent B-splines, placing a Gaussian tensor-product penalty prior on the coefficients. The posterior distribution is obtained by updating the prior using a Whittle-type likelihood approximation of the Wilson-Daubechies-Meyer time-frequency coefficients. Separate regularisation in time and frequency allows non-stationary spectral structure to be inferred. An adaptive coarse-grained likelihood reduces the computational cost while preserving fine resolution around sharp spectral features. We first validate the method using repeated simulations of a locally stationary process with known spectrum, showing lower reconstruction errors and narrower posterior intervals as the sample size increases. We then apply it to a year-long simulated LISA data set with stationary instrumental single link spectra, an evolving time-delay interferometry (TDI) response, and a modulated Galactic foreground. We compare a flexible total power spectral density (PSD) model with models incorporating progressively stronger physical information and examine both continuous and gapped observations. Allowing time variation substantially improves reconstruction relative to a stationary model, while incorporating physical information about the TDI response further improves recovery, particularly near its moving spectral features. The results illustrate the trade-off between spectral flexibility, physical assumptions, and posterior uncertainty in non-stationary LISA noise estimation.
Nihar Mauskar, Sebastian Salazar, Aaron Schein, Wesley Tansey
Fri 13 Nov
stat.AP
pdf html
- reading growth, likely, drugs, burden
- favorites near “Bayesian Semiparametric Model for Sequential Treatment Decisions with Informative Timing” 0.83
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.74
- looks like clinical and biomedical studies
Abstract
Cancer drugs are often evaluated for efficacy in mice before proceeding to human clinical trials. To do this, human tumors are engrafted onto immune-incompetent mice to create a patient-derived xenograft (PDX). Engrafted tumors can be propagated to evaluate multiple drugs, but the time and costs involved in PDX studies limit high throughput drug screens in mice. Here we present GrAFT-PDX, a Bayesian model of PDX drug screens that aims to reduce the PDX screening burden by enabling imputation of missing experiments and early stopping of likely unsuccessful drugs. Unlike conventional survival or growth models, GrAFT-PDX jointly predicts tumor growth and survival outcomes using a learned latent function decomposition. In experiments on a large-scale PDX dataset, GrAFT-PDX outperforms Cox proportional hazards models at predicting time of death, outperforms machine learning baselines at predicting tumor growth, and recovers known gene-treatment relationships without genomic input.
Zhengchi Ma, Kaifeng Zhang, Anru R. Zhang
Fri 13 Nov
stat.MLcs.LGstat.ME
pdf html
- your work near “Valid Inference with Imperfect Synthetic Data” 0.81
- favorites near “Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss” 0.76
- interest Online learning and statistical learning theory 0.77
Abstract
Data imbalance across sensitive groups and class labels is common in classification problems and can contribute to substantial fairness disparities. We study how such imbalance affects Equal Opportunity and how synthetic data augmentation can mitigate the resulting unfairness. We first identify two mechanisms, prevalence imbalance and group-size imbalance, through which changes in the training distribution shift the learned decision boundary and induce disparities in group true positive rates. Motivated by this analysis, we study exact full-balance augmentation, which equalizes all group-label cells. We establish conditions under which full balancing strictly improves Equal Opportunity together with a prediction-error comparison under the oracle balanced distribution, and show that the Equal Opportunity gap of the synthetic-data classifier converges to that of the oracle balanced classifier as the sample size grows and the generator error vanishes. The finite-sample convergence rate agrees, up to logarithmic factors and generator-approximation error, with the information-theoretic lower-bound rate under the corresponding regularity regimes. Since exact balancing need not provide the best fairness-accuracy trade-off in practice, we further propose Accuracy-Constrained Adaptive Synthetic Augmentation (ACASA), which learns both where and how much to augment by minimizing estimated fairness disparity subject to a constraint on predictive accuracy. We provide a finite-sample guarantee comparing the selected augmentation allocation with the population-optimal accuracy-feasible allocation. Simulation studies and real data application support the theoretical findings and demonstrate that synthetic augmentation can substantially reduce fairness disparities while achieving favorable empirical fairness-accuracy trade-offs.
Bowen Gang, Yin Xia
Fri 13 Nov
stat.APstat.ME
pdf html
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.75
- reading mathematics, empirical, gains, mirror
- favorites near “More power to you: Using machine learning to augment human coding for more efficient inference in text-based randomized trials” 0.82
Abstract
Unequal recovery from pandemic learning losses raises concerns about widening achievement gaps between economically disadvantaged students and their peers. This study seeks to identify districts whose mathematics gaps widened in every annual interval from 2022 to 2025. The analysis uses estimates from the Stanford Education Data Archive (SEDA) for 5,600 U.S. school districts. Identifying such districts requires accounting for correlated errors and differing precision in the estimated annual changes, as well as multiplicity. The true annual gap changes in each district form an effect vector, and persistent widening means that every component is positive. To address this class of multiple testing problems, we propose an empirical Bayes mirror procedure. It uses an estimated distribution of effect vectors to guide ranking and covariance-adjusted mirror observations to calibrate selection. The procedure achieves finite-sample false discovery rate (FDR) control under a Gaussian observation model, without requiring accurate estimation of the posterior null probabilities used for ranking. Simulations demonstrate power gains from incorporating information about the distribution of effect vectors. At a target FDR of 25%, the SEDA analysis identifies eight districts in five states with evidence of widening gaps in every annual interval.
Hoang-Viet Nguyen, Hai-Dang Dau, Alexandre H. Thiéry
Wed 18 Nov
stat.MLcs.LG
pdf html
- reading good, sequential, fewer, achieves competitive
- interest Language models as statistical objects 0.75
- your work near “Timing as an Action: Learning When to Observe and Act” 0.72
Abstract
Learning a neural sampler can reduce the cost of repeated sampling from an un-normalized distribution, but obtaining good training data remains difficult. We develop a corrected self-training scheme across a sequence of tempered targets, making use of importance sampling and Markov chain Monte Carlo to correct the samples and replay buffers to store them. We introduce Sequential Score Sampler (SSS) as a particular case. In SSS, the proposal is a score-based diffusion trained by regression on energy gradients; and the samples are corrected by importance weights on the path space. After training, samples can be collected directly from the model, from its importance-corrected proposals, or from the final-temperature buffer. The method amortizes repeated sampling by encoding the global structure of the target in the model. SSS achieves competitive sampling accuracy across the benchmarks and, on several targets, uses shorter runtimes with fewer energy evaluations than competing samplers.
Xu Yang, Haihui Shen
Tue 17 Nov
stat.APstat.CO
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.78
- favorites near “Dynamically Optimal Treatment Allocation” 0.77
- interest Reinforcement learning, policy learning and off-policy evaluation 0.75
Abstract
In service systems, the arrival rate in the same time period and day of the week exhibits randomness, yet can be more accurately estimated or forecast right before that period, which motivates a two-stage staffing scheme with advance base staffing and pre-shift surge staffing. Existing approaches usually impose strong assumptions on the arrival process and service systems' operational logic and other further approximations (i.e., stylized models) to derive explicit staffing policies. This paper proposes a simulation optimization approach (SOA) for more general arrival processes and service systems by effectively addressing some intertwined challenges. By recasting the original formulation into computationally tractable structures and designing an efficient coordinate-search algorithm, it constructs a two-stage staffing lookup table offline, which allows managers to make real-time surge staffing decisions without any further simulation. Comprehensive numerical experiments demonstrate that the proposed SOA has significantly better staffing performance and substantially broader applicability compared to existing approaches. Even under the stylized models, the cost savings delivered by our SOA compared with existing approaches range from the low tens to several tens of percent, depending on different problem parameters; the advantage further amplifies when the actual problem deviates from the stylized models. In a nurse-staffing case study grounded in real-world data, the proposed SOA delivers approximately 8% to 13% cost savings over a state-of-the-art benchmark.
Ali Nikseresht
Tue 17 Nov
eess.SYcs.SYstat.APstat.ME
pdf html
- reading long memory, reserves, long, makes
- interest Computational methods for heterogeneous agent and dynamic economic models 0.73
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.70
Abstract
Day-ahead load forecasts set balancing reserves, including the energy a pumped-storage plant holds against forecast errors. Both depend on three effects that are usually modeled separately, namely the thermal inertia of buildings, which makes load lag temperature, structural breaks, which shift demand, and the long memory of forecast errors, which keeps them from cancelling out. We prove that averaging building responses over Cole-Cole distributed time constants yields a fractional temperature kernel with nonnegative weights exactly when its order is at most one. A Bayesian information criterion that ignores long memory selects spurious breaks with probability tending to one, so breaks and memory are estimated jointly, and an online break detector keeps its false-alarm rate below any prescribed level. Long memory also makes the required storage energy grow faster than the square root of the horizon, so a second-order cone program schedules reserves from the joint distribution of hourly errors. On New England and Mid-Atlantic data including the 2020 pandemic, the proposed framework improves on the best of ten benchmarks by 15.7% and 10.3% in mean absolute percentage error. Its reserves breach the storage energy limit on 4.1% and 8.5% of days, against 36.7% and 57.5% with error correlations ignored.
Sicheng Shu, Parker Knight, Huitao Li, Yuan Luo, Yifan Peng, Shawn N. Murphy, et al.
Tue 17 Nov
cs.AIstat.AP
pdf html
- reading did, empirical, evidence, considers
- your work near “Timing as an Action: Learning When to Observe and Act” 0.75
- favorites near “Underspecification Presents Challenges for Credibility in Modern Machine Learning” 0.81
Abstract
Patient information can guide electronic health record (EHR) generation, but how its amount shapes predictive utility and disclosure risk remains poorly characterized. Here we present EHRDiB, which unifies visit completion and synthesis by distinguishing original missingness from withheld inputs and combining a shared multimodal representation with an evidence-conditioned latent diffusion bridge. We compared EHRDiB with three completion and eight synthesis methods on MIMIC-IV and numerical and discretized Duke Health representations. Evaluation covered downstream predictive utility, generative diversity, and distributional fidelity for both tasks, and disclosure risk for synthesis. EHRDiB achieved the smallest utility gaps from real-data-trained classifiers in seven of eight completion comparisons. In synthesis, AUPRC gaps were 45.2% smaller for ten MIMIC-IV principal-diagnosis targets and 53.3% smaller for numerical Duke Health data relative to the best comparator. Synthesis with calibrated thresholding yielded the lowest marginal distribution error across all three datasets, with Coverage above 0.95 for both output rules. Across twelve evidence-fraction intervals in MIMIC-IV synthesis, membership-inference AUROC increased from 0.516 to 0.598 as the mean target evidence fraction rose from 0.20 to 0.50; utility gaps narrowed relative to the lowest evidence configuration but did not decrease monotonically. Our study provides a unified framework and empirical foundation for EHR generation that considers both its utility for clinical analysis and its information disclosure risk.
Hoang Pham Cong, Long D. Nguyen, Ho Viet Duc Luong, Nguyen Tran Minh Nhat, Binh P. Nguyen, Huynh Thi Thanh Binh
Thu 19 Nov
cs.AIstat.ML
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.78
- interest Language models as statistical objects 0.74
- favorites near “Underspecification Presents Challenges for Credibility in Modern Machine Learning” 0.82
Abstract
A deployed multimodal predictor may face missing modalities after its weights can no longer be changed. This raises a question that training-time repair does not answer: what part of the resulting error is still correctable from the information that remains? We show that this part is characterized by the Bayes gap, the difference between the model's prediction and the Bayes prediction given the observed modalities. For cross-entropy and squared error, excess risk measures this gap in the geometry of the loss; ambiguity in the observed modalities cannot be removed. Thus, a badly degraded prediction is not necessarily a correctable one. Estimating the Bayes prediction normally requires labels. However, when a model observing more modalities is conditionally correct, averaging its outputs conditional on the remaining modalities recovers this target without retrieval labels; with the same fixed blend and neighbors, it is no worse in expected cross-entropy or Brier risk than averaging hard labels. Some such assumption is unavoidable: otherwise the target is not identifiable from unlabeled data alone. GUARD (Gated Utility-Aware Recalibration for Deployment) leaves the deployed predictor unchanged. It estimates a candidate correction by retrieval, learns from deployment-matched labeled data when correction is likely to help, and calibrates a threshold on a held-out split by conformal risk control. Under exchangeability, the marginal probability that GUARD applies a correction whose loss increase exceeds a prescribed tolerance is bounded by a preset level, regardless of the learned score. In controlled experiments, correction gain follows headroom rather than raw error; across nine frozen model-dataset pairs spanning six domains, GUARD retains most of blanket correction's gain within harm budgets that blanket correction often exceeds.
Namrata Datta, Victor Miranda Soberanis
Fri 13 Nov
stat.APcs.LG
pdf html
- reading versus, inflation, probe, future
- interest Language models as statistical objects 0.77
- your work near “Timing as an Action: Learning When to Observe and Act” 0.74
Abstract
Reported accuracy in Parkinson's disease (PD) machine learning can reflect the evaluation protocol rather than real signal. We ask how much evaluation choices inflate performance by comparing naive and rigorous protocols on voice, telemonitoring and blood-transcriptomic datasets. Subject-aware splitting reduced voice AUROC from 0.892 to 0.780 and telemonitoring R^2 from 0.105 to -0.037; Random Forest recovered neither gap. Supervised probe selection before splitting inflated transcriptomic AUROC from 0.711 to 0.998. A nested 50-probe panel matched a 1,000-probe panel (0.811 versus 0.844), and recruitment-wave batch structure was detected. Future work should validate a stable panel on an independent cohort.
Bolun Liu, Vamsi Manthena, Wally Toh, Wenjing Zheng
Tue 17 Nov
stat.APstat.ME
pdf html
- reading game, games, causal, nominal
- favorites near “Recent Developments in Machine Learning Methods for Stochastic Control and Games” 0.74
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.70
Abstract
At Roblox, viral hits can change how exposed players engage with the viral game and with the rest of the platform. We develop a causal mediation framework to decompose the effect of viral-game playtime on subsequent player engagement into a direct effect and indirect effects through measured player interaction and discovery behaviors. The framework uses a causal DAG to state the assumed causal structure and a time-ordered design that measures covariates, exposure, mediators, and outcomes in non-overlapping time windows. Under stated identification assumptions and a model linear in the exposure and each mediator at fixed covariates, we estimate every path as a function of player covariates with cross-fitted R-learners, which use only prediction models and therefore scale to platform-size data. Averaging the fitted surfaces inherits their regularization bias, so we add a per-player bias correction before averaging (Chernozhukov et al., 2018). In simulation, the correction restored nominal coverage of 90% to 95%, against at most 39% for intervals built by averaging the surfaces. Counterintuitively, interval coverage worsened with more data on a difficult mediator-outcome path, bias shrank more slowly than the standard error, reducing corrected coverage from 93.6% to 57.3% as n increased from 2k to 200k. We applied the same design to two de-identified viral games with different lifecycle histories, comparing effects across outcome scopes, player subgroups, and a placebo window without a viral event. The results indicate that subsequent engagement is not explained by viral exposure alone, player interaction behaviors mediate 46% engagement within the viral game, whereas new game discovery mediates 52% engagement in other games. Finally, we compare the estimator with its alternatives and generalize the study design into a reusable framework for similar mediation studies across product domains.
Yulia Alexandr, Moritz Grillo, Thiago Holleben, Guido Montúfar, Teresa Yu
Thu 19 Nov
math.AGstat.ML
pdf html
- reading equations, proof, algebraic, constraints
- favorites near “Why Does Deep Learning Work? - A Perspective From Group Theory” 0.75
- interest Computational methods for heterogeneous agent and dynamic economic models 0.71
Abstract
We give a complete algebraic characterization of two-block shallow ReLU pattern varieties. We show that the determinantal equations arising from rank constraints imposed by the activation patterns generate the defining ideals, thereby proving a conjecture of Alexandr-Mont\'ufar. The set-theoretic proof uses a characterization of orbit closures of matrix pencils in terms of the Kronecker canonical form.
Haisheng Fu, Jie Liang, Runyu Yang, Feng Ding, Xin Dong, Zixiao Jin, et al.
Thu 19 Nov
eess.IVcs.CVstat.AP
pdf html
- reading gains, curves, good, future
- favorites near “Data Science at the Singularity” 0.81
- interest Language models as statistical objects 0.74
Abstract
Learned image compression (LIC) papers often report BD-rate improvements of only a few percent. They are measured against VTM reference software whose version keeps changing: the claims we audit name 7 versions, 5 in 2025 papers alone. But how comparable are these numbers across papers, and what evidence should future papers give for such small gains? We introduce LIC-Audit, an analysis of 353 released rate-distortion curves from 36 repositories, and a controlled re-evaluation of eight public codec releases with 28,193 encode-decode records. We identify three problems. First, different VTM reference runs change BD-rates by up to 3.56 percentage points, and on CLIC Pro published numbers order 10 of 55 method pairs differently from a common computation; comparing codecs directly over a fixed quality range removes this dependence. Second, releases use different and mostly undocumented conventions for counting bits, computing PSNR, and padding images. Under consistent measurement, the GLIC method's 1.00% lead over the HPCM method on Tecnick becomes a 1.43% deficit, changing only the padding reverses the comparison of HPCM and CMIC, and most gaps narrow when MS-SSIM replaces PSNR. Third, average curves hide how much results vary from image to image, so whether a small gain is statistically significant depends on the assumed variation. An optimistic assumption calls 12.0% of comparisons between equally good codecs significant on test sets as small as Kodak, instead of 5%; a cautious one leaves 18 of 45 differences between recent codecs undecided, yet per-image results show 12 of them to be significant. Beyond auditing past results, LIC-Audit provides a versioned evaluation protocol, a public leaderboard website with a scoring tool, source-linked benchmark data, per-image measurements, and reporting guidelines, to help future LIC papers make comparable and verifiable claims of progress.
Ruizhi Deng, Guohao Shen, Tangrui Xiong
Fri 13 Nov
cs.LGmath.STstat.MLstat.TH
pdf html
- favorites near “Why Does Deep Learning Work? - A Perspective From Group Theory” 0.79
- interest Computational methods for heterogeneous agent and dynamic economic models 0.75
- your work near “Perturbation Methods for Incomplete Markets Economies” 0.74
Abstract
In neural networks, distinct parameter configurations often realize the same input--output mapping, indicating that nominal parameter counts can overstate functional complexity. In this paper, we characterize the parameter-space symmetries of ReLU and Leaky ReLU networks across fully connected, convolutional, and residual architectures. We prove that every quasianalytic parameter map preserving the network function is a composition of positive channel scaling and channel permutation. By quotienting out both symmetries, we derive uniform covering-number upper bounds in which positive scaling reduces the complexity exponent from the total parameter count $S$ to $S-H$ (with $H$ denoting the number of scalable channels) and permutation symmetry divides the bound by the factorial count $G$ of channel permutations. We establish matching lower bounds showing that the exponent $S-H$ is sharp in general and that the divisor $G$ is attained for one-hidden-layer and generically identifiable networks. We further extend these results to general continuous piecewise-linear activated networks. Finally, we apply our covering-number bounds to derive statistical risk guarantees for empirical risk minimization and capacity control beyond i.i.d.\ settings, and show how the underlying symmetries help explain weight rebalancing in function approximation and flat directions in empirical risk optimization.
Taimoor Ahmad
Fri 13 Nov
stat.AP
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.76
- interest Causal inference and identification in econometrics 0.79
- reading state, forecasting, nominal, tuned
Abstract
SaaS capability assessments can become stale between questionnaires, while delivery telemetry offers incomplete and correlated evidence. Continuous assessment and Bayesian delivery analysis already exist, leaving uncertainty about how biased surveys and missing observations should influence capability estimates.We implement SAAS-MATURITY-TWIN, an anchored Gaussian state model that jointly filters capability proxies and persistent survey offsets. Source covariance prevents correlated observations from being counted independently. Conditional uncertainty supports forecasting and budgeted intervention screening, with action effects estimated from randomized synthetic cohorts. Ten seeds generate forty paired worlds spanning reference observations, extensive missingness, informative missingness, and increased noise. Reference latent root mean squared error is 0.1261, compared with 0.1452 for a tuned moving average; nominal ninety percent intervals cover 90.3 percent. Improved state recovery produces only small incident forecasting differences. Adaptive conformal inference and an adapted conformal PID controller are evaluated separately for observed telemetry, where noise shifts expose calibration limitations. A licensed TravisTorrent pilot evaluates build prediction on 13 074 held out builds from 224 projects. Logistic regression achieves a Brier score of 0.1289, close to random forests. These results support reproducible methodological testing, but do not establish real organizational maturity, intervention effectiveness, or production readiness. Independent calibration and longitudinal validation remain necessary.
Amadou Diadie Ba
Tue 17 Nov
math.STstat.TH
pdf html
- interest Adaptive data collection and anytime-valid inference 0.82
- your work near “Perturbation Methods for Incomplete Markets Economies” 0.76
- favorites near “Applied Koopmanism” 0.80
Abstract
We develop a complete asymptotic theory for monitoring data drift in production artificial intelligence (AI) systems whose inputs take values in a finite set of categories. The framework combines two complementary inferential tools. Under the no-drift hypothesis $H_0$, the two-sample likelihood-ratio statistic $G$ of Wilks follows a universal Chi_{r-1}^2 law, independently of the divergence chosen and of the sample sizes; the calibrated Population Stability Index (PSI) of Yurdakul and Naranjo (2021) shares the same Chi_{r-1}^2 limit and is asymptotically equivalent to the $G$-test under $H_0$. Under the drift hypothesis $H_1$, the plug-in phi-divergence is asymptotically normal with a variance computed in closed form for the one-sample, two-sample and symmetrized configurations, yielding standardized $Z$-statistics, detection boundaries, power functions and sample-size planning rules. We supply a per-category decomposition that localizes the drift on specific categories, a Benjamini--Hochberg procedure for monitoring many features simultaneously, and an explicit comparison with the heuristic PSI cut-offs of industrial practice. Two fully worked simulation examples - one one-sample with the R\'enyi divergence and multiple testing, one two-sample with the Kullback-Leibler (KL) divergence and sample-size planning - illustrate every notion of the paper on realistic production data.
Yegor Baranovski
Wed 18 Nov
stat.MEstat.ML
pdf html
- interest Adaptive data collection and anytime-valid inference 0.81
- your work near “Valid Inference with Imperfect Synthetic Data” 0.75
- reading earnings, best, finite-sample bounds, drift estimation
Abstract
Cross-validation weights all folds equally, so under drift the oldest folds count as much as the newest. We model drift by a H\"older condition in total variation distance and give finite-sample bounds on the excess deployment risk of the configuration that sliding-window cross-validation selects. Each bound splits into drift, estimation, and validation-noise terms. With equal weights the drift terms saturate as the record grows. With fold weights fixed in advance, the drift terms become weighted averages, which shrink as weight moves toward deployment. The noise term grows as the weights concentrate and shrinks with the batch size. The drift terms are linear in the weights and the noise term is convex, so minimizing the bound over the weights is a convex problem. Applications to U.S. earnings and flight delays show both sides of this trade-off. Weight on recent folds pays off when a change in the best predictor persists and each period has many observations, and equal weights do best when changes are short-lived.
Mirkan Emir Sancak
Fri 13 Nov
stat.APcs.CE
pdf html
- favorites near “Monte Carlo Geometry Processing” 0.82
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.71
- interest Empirical Bayes, shrinkage and decision theory 0.75
Abstract
Sampling geometry may influence both reconstruction stability and the smoothing needed to control noise. This study distinguishes geometry-dependent failure of exact multiquadric radial basis function (RBF) interpolation from variation in validation-selected smoothing. A synthetic benchmark crossed four response families, four 16-point geometries, three slice levels, five observation regimes, and 24 repeated holdouts, yielding 17,280 method-level evaluations. Clough--Tocher cubic interpolation, exact RBF, and smoothed RBF were compared, followed by a Water Quality Portal pilot with 1,950 complete activities. On the synthetic shared-valid domain, exact RBF achieved the lowest noise-free mean RMSE (0.0206), whereas smoothed RBF performed best under noise (0.1807 versus 0.2414 for cubic and 0.6106 for exact RBF). Clustered-gap support produced the largest noisy exact-RBF RMSE (1.2706). Geometry added little explanatory power for log-transformed smoothing: the combined model had $R^2=0.4259$, compared with 0.4096 for noise alone. Reported failure-regression results suggest a stronger geometry signal, but remain unverified without their original model specification. In the environmental pilot, smoothed RBF achieved mean RMSE 0.9877 with no negative-$R^2$ failures. Conflicting responses at identical predictor coordinates made exact interpolation unsuitable. All environmental smoothed-RBF fits selected the largest candidate, preventing geometry--smoothing correlation analysis. The findings motivate geometry-aware failure screening but do not establish a transferable rule for smoothing magnitude. Independent multisite validation, appropriate coordinate scaling, joint scale--smoothing tuning, and conditioning diagnostics are needed before an adaptive policy can be claimed.
Kuancheng Ye
Fri 13 Nov
stat.MLcs.CLcs.CRcs.LG
pdf html
- reading document, documents, confidence sequences, make
- your work near “Perturbation Methods for Incomplete Markets Economies” 0.73
- favorites near “Adapting Text Embeddings for Causal Inference” 0.76
Abstract
Watermark detectors decide whether one document came from a watermarked language model. Auditors often need a different number: the share of a corpus that did. We study this share for Gumbel-max and green-list (KGW) watermarks when every document may have been edited by an unknown amount. Without restrictions on the edits, the upper end of the identified set is 1. Combining a known least-favourable next-token distribution with a sequential coupling, we derive a power envelope indexed by three budgets: a retention floor, an entropy budget, and a budget for in-place substitutions, which for Gumbel-max make edited positions look less watermarked than human text. The envelope gives an identified set that is sharp for the alternatives it dominates, finite-sample bounds on the realized share of a fixed corpus with either the exact null or a null calibrated on an independent human sample, and confidence sequences for streams. In Gumbel-max simulations where the envelope binds, the bounds cover, and the threshold that maximizes the gap between envelope and null is generally as tight as twelve thresholds. On text from OPT-1.3B and Qwen3-1.7B, 12,000 fresh human documents scored under three keys show no robust departure from the exact null, and all bounds covered in resampling trials from finite pools of documents with random edits, substitutions and paraphrases; at a share of 0.2 the exact-null bound is about [0.14,0.38]. These results are fragile in two ways we make precise. LLM-paraphrased KGW pools fall below the envelope, yet at 2000 documents the bounds still covered in these trials. A simulated null error of 0.01, comparable to the largest deviation we observed at our thresholds, is likewise harmless at 2000 documents but breaks exact-null coverage at 100,000; a null calibrated on an independent human sample keeps coverage there, but needs a large reference sample to stay narrow.
Kai-Wayne Liu, Ryan Tian En Pang, Pholpat Durongbhan
Tue 17 Nov
q-bio.QMq-bio.NCstat.ME
pdf html
- reading apparent, choices, inference using, empirical
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.75
- interest Empirical Bayes, shrinkage and decision theory 0.74
Abstract
Background: The weighted rich-club coefficient tests whether a network's most strongly connected nodes are preferentially interconnected. Its interpretation depends on two often-implicit choices: how the randomised comparison networks are constructed, and whether the hubs are re-identified in each comparison network or held fixed at those of the observed network. When hubs are defined by connection strength, these choices change what the coefficient tests. New method: We examined how these choices affect inference using resting-state EEG from 20 people with Alzheimer's disease and 20 controls. Holding the data fixed, we compared four analysis pipelines, crossing two ways of building the comparison networks (global weight reshuffle or strength-conditioned null) with the two hub choices, and ran a calibration study using data constructed to contain no preferential hub coupling beyond nodal strength. Results: An apparent eyes-open group difference in rich-club organisation appeared only under the pipeline combining a global weight reshuffle with re-identified hubs. Under the empirical resampling ensemble, that pipeline reported a group difference in 43% of pseudo-studies, always higher in the Alzheimer's disease group, while the other three stayed near the expected 5%. Comparison with existing methods: Unlike global weight reshuffling, strength-conditioned nulls approximately preserve nodal strengths; fixing observed hubs also retains the identity of the core being tested. Conclusions: The same data can support or dissolve an apparent rich-club difference depending on these choices. No single choice is universally correct, but reporting which were made, and why they fit the question being asked, is essential to interpreting the result.
Lorenzo Mauri, Samantha Gleich, Michael Johnson, Andy Liaw, Daniel Skomski, Federico Ferrari
Fri 13 Nov
stat.AP
pdf html
- reading equation, rates, nonlinear, special
- interest Kernels, operators and functional data 0.76
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.73
Abstract
Predictive stability modeling is widely used in pharmaceutical development to establish product shelf-life by identifying the kinetic parameters that govern degradation behavior and using these parameters to extrapolate out in time. Recent approaches typically assume linear degradation dynamics and rely on the Arrhenius equation to model temperature-specific reaction rates. However, observed degradation profiles may reflect nonlinear reaction dynamics and more complex temperature-specific reaction rates, potentially leading to reduced accuracy in long-term predictions and motivating a more flexible approach. We propose a novel Bayesian framework which extends predictive stability modeling beyond simple Arrhenius low-order kinetic dynamics and accommodates complex degradation mechanisms. The degradation process is modeled through the Sestak-Berggren kinetic formulation, a flexible ordinary differential equation encompassing many classical degradation models as special cases. To facilitate model interpretability and model exploration, probabilistic model selection on kinetic parameters automatically determines whether nonlinear reaction components are supported by the data, allowing the model to adaptively collapse to simpler dynamics when appropriate. To address departure from classical Arrhenius kinetics, we develop a generalized temperature reaction rate formulation in which temperature-specific reaction rates are shrunk toward Arrhenius predictions through a hierarchical shrinkage prior while allowing departures when supported by the data. This approach yields robustness to misspecification and reduces sensitivity to potentially influential high-temperature observations commonly encountered in accelerated stability experiments. The resulting framework provides a principled and flexible statistical tool for model-informed shelf-life determination.
Tiantian Yang, Dongwei Chen
Thu 19 Nov
stat.MEmath.PR
pdf html
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.73
- interest Macroeconometrics and time series 0.75
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.71
Abstract
The target diagram is often used in geosciences, which provides a visualization tool for model evaluation and shows the temporal or spatial matching of data. However, its construction requires paired data of equal sample sizes, limiting its application to model comparisons involving unpaired samples or probability distributions. In this study, we propose the Wasserstein target diagram based on the 2-Wasserstein distance from optimal transport. Unlike the original target diagram, the Wasserstein target diagram does not require paired data of equal sizes and evaluates models from the probability distribution perspective. We further establish that the two target diagrams coincide when the given paired data are comonotone, which is common in climate science, such as cumulative precipitation and solar radiation. Through two simulation studies and an application to wind speed distribution fitting, we demonstrate that the proposed diagram can detect the probability distribution discrepancies that the original target diagram may not distinguish. The Wasserstein target diagram therefore lifts the model evaluation to the probability level and provides a complementary approach to estimating the agreement in probability distributions.
Yanhang Zhang, Jingfu Peng, Yuhong Yang
Tue 17 Nov
math.STstat.TH
pdf html
- interest Online learning and statistical learning theory 0.83
- favorites near “A U-turn on Double Descent: Rethinking Parameter Counting in Statistical Learning” 0.83
- your work near “Timing as an Action: Learning When to Observe and Act” 0.70
Abstract
Unlike in continuous regression, a small input perturbation in classification can move an input across a decision boundary and flip the predicted label. Decision-boundary geometry is therefore essential to understanding adversarial classification risk. Under anisotropic smoothness and margin conditions, we quantify this geometric contribution and establish sharp minimax rates for adversarial excess risk relative to the ordinary Bayes risk. These rates comprise a statistical term governed by the harmonic mean smoothness and a geometric cost of adversarial perturbations governed by the least smooth direction. As smoothness and the margin exponent vary, the adversarial term exhibits a phase transition from a fractional power of the attack radius to linear scaling, with a logarithmic correction in a critical regime. In addition, we construct an adaptive classifier that attains these rates over a range of smoothness and margin classes. We further characterize the minimax rate for adversarial excess risk under a budget on ordinary excess risk, yielding an explicit trade-off curve. The curve identifies attack regimes where ordinary minimax optimality entails a suboptimal adversarial rate. Finally, numerical experiments support the theoretical findings.
Chien-Chung Wang, Benjamin A. Shaby
Thu 19 Nov
stat.AP
pdf html
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.72
- reading atlantic, like, long, south
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.68
Abstract
We propose a model for spatial extremes what realistically accommodates nonconvex domains. Most existing models assume that tail dependence between locations depends only on their separation. Even if extended to be anisotropic or nonstationary, they still cannot accommodate features between points that may mediate the dependence. This assumption limits their applicability in settings where physical barriers like mountains, coastlines, or islands may attenuate or even block dependence in extreme events. In this work, we use the idea of a visibility graph to modify an existing spatial random scale model. This results in a tail dependence model that naturally accommodates nonconvex spatial domains, where the nonconvex features are interpreted as physical barriers. Furthermore, the model inherits key properties from the scale mixture like the ability to represent asymptotic independence at long ranges and either asymptotic dependence or asymptotic independence at short ranges. We apply our model to two applications in which physical barriers feature prominently in determining the spatial structure of extreme events. The first is precipitation in the South Island of New Zealand, where the Southern Alps shelter the eastern region from the effects of atmospheric rivers which arrive from the west. The second is significant wave heights in the North Atlantic, where the Faroe Islands tend to block the effects of extratropical cyclones traveling into the North Sea.
Charles-Albert Lehalle, Juliette Anglade, Mario Massy
Tue 17 Nov
stat.ME
pdf html
- reading narratives, narrative, empirical, explains
- favorites near “Adapting Text Embeddings for Causal Inference” 0.82
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.66
Abstract
We propose an algebraic methodology for measuring the relative prevalence of competing narratives within a time-evolving corpus of texts. Standard cosine similarity between article embedding and narrative vectors fails to isolate the idiosyncratic contribution of each narrative, because narratives sharing a common topic produce strongly correlated vector representations. We address this by orthogonalizing the narrative vectors via a global max-min optimization on the Stiefel manifold, which neutralizes shared semantic components while preserving each narrative's specific direction. We ground this geometric procedure in a linear generative model for article embedding that provides an unbiased estimator of narrative prevalence and formally explains why the naive approach suffers from a correlation bias. Empirical validation on the publicly available News Category Dataset (Huff Post, 2012-2018) confirms that the orthogonalized estimator recovers narrative-specific dynamics aligned with known historical events, whereas the naive cosine projection yields synchronized, uninformative time series.
Rafael Mouallem Rosa, Julyan Arbel
Thu 19 Nov
cs.LGstat.ML
pdf html
- favorites near “PAC-Bayes Compression Bounds So Tight That They Can Explain Generalization” 0.84
- interest Language models as statistical objects 0.76
- reading large, resulting model, reasonable, generally
Abstract
Highly parameterized convolutional neural networks often achieve strong predictive performance, but their memory, computation, and deployment costs motivate the search for methods that recover compact architectures from large reference models. We propose SoU-RHS, a Bayesian structured-sparsity procedure that augments Regularized Horseshoe Shrinkage (RHS) with the so-called Sources of Uncertainty Sparse Global-Local mechanism (SoU-SGL). The resulting model sharpens the separation between active and inactive filters, providing an effective mechanism for architecture recovery. A key property of SoU-RHS is stability: across different overparameterized starting widths and under a reasonable range of pruning thresholds, it recovers compact architectures with similar total size and layer-wise allocation. This supports the interpretation that the procedure is identifying a compact structure adapted to the data, rather than merely compressing each initial model separately. Across standard image-classification benchmarks, SoU-RHS recovers substantially smaller architectures than deterministic structured-sparsity baselines, while maintaining competitive accuracy, achieving lower negative log-likelihood in every evaluated configuration and generally lower expected calibration error.
Jilang Miao, Guillaume L. Giudicelli, William Reed Kendrick
Tue 17 Nov
physics.comp-phstat.CO
pdf html
- reading theory, operator, moment, gives
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.73
- interest Bayesian computation and structural estimation 0.78
Abstract
Predicting correlations between source generations can support uncertainty estimation and variance reduction in Monte Carlo transport. We derive a two-dimensional benchmark for planar isotropic exponential flights in a reflecting square. An image construction gives the exact collision and absorption spectra. Critical branching preserves the mean transport operator but changes mode covariances through the offspring second moment. These covariances give tally autocorrelations and integrated autocorrelation times, with a large-population, finite-window approximation for normalized source tallies. Direct Monte Carlo measurements test the predictions for absorption/rebirth and two-offspring fission. The benchmark provides a reference for assessing reduced correlation models, for correcting reported tally uncertainties, and for checking the tallies of a production code for correlation bias.
Taeseok Park, Jesús Arroyo
Mon 16 Nov
stat.ME
pdf html
- interest Kernels, operators and functional data 0.78
- favorites near “Hypoelliptic diffusion maps I: tangent bundles” 0.82
- your work near “Estimating Treatment Effects with Observed Confounders and Mediators” 0.77
Abstract
Multilayer network data, arising from multiple sources, repeated measurements, or multi-view interactions, are increasingly common, highlighting the need for statistical methods that jointly leverage information across layers while allowing for layer heterogeneity and computational scalability. Here, we study a multilayer degree-corrected mixed membership model (ML-DCMM), in which nodes share a common mixed membership across layers while degree effects and block connectivity patterns vary by layer. This model encompasses several existing models for multilayer data while preserving identifiability and interpretation. We develop a spectral estimation procedure that integrates layer-wise embeddings with corner estimation for membership recovery. A further refinement that connects the resulting representation to a ParaTuck-2 tensor decomposition is also introduced. We establish theoretical guarantees for the proposed spectral method, showing that the probability of near-corner recovery increases with the number of layers, and we provide the convergence rate for the resulting membership estimator. We establish a minimax lower bound for membership estimation, which does not improve with increasing number of layers. Simulation studies show that ML-DCMM can outperform other approaches under strong layer heterogeneity. In an analysis of London bike-sharing networks, the methodology recovers interpretable memberships and achieves lower held-out edge prediction error under edge-sampling cross-validation than competing models.
Yasaman Mohammadpour, Pouya Shaeri
Tue 17 Nov
cs.LGcs.AIeess.SPstat.ML
pdf html
- favorites near “Spectral State Space Models” 0.75
- reading versus, pretraining, gains, come
- interest Language models as statistical objects 0.73
Abstract
Self-supervised learning for wearable sensing is usually confined to a single dataset or device family, leaving open whether representations transfer across heterogeneous sensor configurations and whether gains come from pretraining rather than architecture or richer sensor input. We present HeteroWear, a sensor-aware self-supervised framework pretrained on an eight-dataset wearable corpus (173 subjects, 358,261 windows) spanning inertial and physiological signals across body locations, devices, and sampling rates. HeteroWear separates within-sensor temporal modeling from cross-sensor relational modeling using sensor-context embeddings, temporal and cross-sensor Transformers, and latent cross-attention fusion. Training combines structured sensor dropout, cross-configuration consistency, and per-axis masked sensor modeling; missing sensors are omitted rather than reconstructed. Under leave-one-dataset-out pretraining and subject-disjoint evaluation on PAMAP2, WISDM, HHAR, HARTH, and UCI-HAR, HeteroWear improves mean macro-F1 from 0.688 to 0.724 over the identical architecture trained from scratch, with the largest gain on HHAR (0.823 versus 0.741), and achieves the highest macro-F1 on four of five benchmarks relative to single-accelerometer supervised baselines. Held-out-device evaluation yields a mean +0.066 macro-F1 improvement. Under 50% test-time channel removal, HeteroWear retains 73.8% of full-channel performance versus 64.1% for the same architecture trained from scratch without sensor dropout. Ablations show that structured sensor dropout primarily drives robustness, while cross-configuration consistency and sensor-context encoding contribute to transfer.
Chong Gan, Jacqueline May, Zeny Feng
Fri 13 Nov
stat.ME
pdf html
- favorites near “Distributional Random Forests: Heterogeneity Adjustment and Multivariate Distributional Regression” 0.85
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.68
- interest Macroeconometrics and time series 0.74
Abstract
Finite mixture regression models are widely used to capture the heterogeneous effects of covariates on a response variable. Our study, motivated by the analysis of Chiroptera, explores the dynamic effects of various environmental, biological, and human-related covariates on the evolutionary development of bat forearm using finite Gaussian mixture regression (GMR) models. In this paper, we address two key questions: (1) which covariates are associated with the variation in forearm length; and (2) whether the associated covariates have the same effect across all subpopulations. To answer these questions, we propose two stepwise procedures within the finite GMR framework based on the weighted significance test (WEST): (1) a stepwise variable selection procedure to identify relevant influential covariates, and (2) a stepwise effect-type determination procedure to determine each relevant covariate as having either a homogeneous or heterogeneous effect across subpopulations. Based on these procedures, we introduce a stepwise model selection procedure for finite mixture regression, illustrated through finite GMR models. Simulation studies are conducted to assess the performances of our proposed procedures. The proposed procedure will be applied to the Chiroptera dataset.
Gorana Mijatovic, Marta Iovino, Michal Javorka, Luca Faes
Fri 13 Nov
stat.APphysics.data-an
pdf html
- reading causal, consistent, dynamic, signal processing
- interest Macroeconometrics and time series 0.77
- favorites near “Eigendecompositions of Transfer Operators in Reproducing Kernel Hilbert Spaces” 0.79
Abstract
The quantification of dynamic interactions in coupled physiological time series is a central question in biomedical signal processing. A critical issue is the presence of instantaneous dependencies, often neglected in conventional Granger causality (GC) analyses despite potentially reflecting physiologically meaningful effects. We investigate time- and frequency-domain measures of causality within the linear autoregressive (AR) framework, treating instantaneous dependencies according to three formulations: (i) the classical strictly causal AR model, providing spectral measures of time-lagged GC and instantaneous interaction; (ii) the Geweke approach, assigning instantaneous dependencies to the direction opposite to that under investigation; (iii) a single extended AR model incorporating causally-oriented instantaneous dependencies, yielding spectral GC measures accounting for both instantaneous and time-lagged causality. The three approaches are evaluated in simulated Gaussian systems, where dynamic measures can be derived analytically and compared with the expected coupling structure. The single extended AR model emerges as the most consistent formulation when instantaneous dependencies have a physiologically meaningful causal direction and is then applied to cardiovascular interactions between heart period (HP) and systolic (SAP) or diastolic arterial pressure (DAP) in healthy subjects during postural and mental stress. Accounting for instantaneous dependencies substantially affects HP-SAP and HP-DAP interactions, enabling better quantification of baroreflex and cardiac run-off mechanisms and characterization of their responses to stress and of low- and high-frequency dynamics. Overall, the framework provides consistent representation of causal interactions across time and frequency and highlights the importance of accounting for instantaneous dependencies in physiological coupling.
Soham Das, Yang Chen, David A. van Dyk, Herman L. Marshall, Vinay L. Kashyap, Michael Smith
Mon 16 Nov
astro-ph.IMstat.AP
pdf html
- reading instruments, hierarchical model, inferences, structural
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.67
- favorites near “Inference with Hamiltonian Sequential Monte Carlo Simulators” 0.69
Abstract
The ongoing calibration of space-based instruments is essential to ensure accurate measurements of physical quantities such as photon flux. The goal is to obtain a unified photon-flux measurement of a single astronomical source using data obtained with multiple instruments. Calibration involves in-flight adjustments to address discrepancies arising from instrument-specific variations. Taking a statistical approach, we extend a previously proposed additive log-Normal hierarchical model with half-variance correction based on a multiplicative physical model for photon counts. The newly proposed Bayesian hierarchical model can leverage domain-specific priors to account for correlated passbands and integrate information across different instruments, sources, and passbands effectively. The model is fitted using Markov chain Monte Carlo and can provide uncertainty quantification for instrument- and source-specific parameter estimation. Moreover, we obtain valid inferences for other structural parameters of the physics model. We use data derived from Chandra and XMM-Newton X-ray observatories to illustrate the proposed inference procedure.
Vanessa McNealis, Erica E. M. Moodie, Alain Fourmigue, Milada Dvorakova, Joseph Cox
Tue 17 Nov
stat.APstat.ME
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.78
- reading did, engage, men, involved
- interest Causal inference and identification in econometrics 0.72
Abstract
Respondent-driven sampling (RDS) is widely used to recruit populations lacking conventional sampling frames. Because RDS oversamples well-connected individuals, weights are commonly applied to improve representativeness. Although methods for cross-sectional RDS analyses are well developed, guidance for longitudinal studies is limited, despite additional selection bias from missed visits and dropout. We develop a framework that jointly accounts for unequal inclusion probabilities at recruitment and attrition during follow-up. The approach combines baseline RDS weights with inverse-probability-of-non-response weights estimated from observed follow-up histories, producing visit-specific weights for estimating prevalences, means, and regression parameters. We clarify when the RDS design should inform attendance models rather than being addressed only during outcome analysis. Using simulations of RDS recruitment, we assess prevalence and standardized prevalence differences under varying recruitment and attrition mechanisms, focusing on bias, variance, and confidence interval coverage. Weighting the response model by baseline RDS weights was important when an unmeasured determinant of degree also predicted response, even when it did not predict the outcome. We illustrate the framework using data from the Montreal site of Engage, a multisite RDS cohort of sexual minority men in three Canadian cities, from 2017-2023. Follow-up involved intermittent missed visits, attrition, and terminal dropout that was more common among participants with smaller networks. Our findings provide practical guidance for combining design-based and attrition adjustments to support valid inference in longitudinal RDS studies.
Pascal Janetzky, Michael Klar, Manuel Wenzel, Tobias Schlagenhauf, Stefan Feuerriegel
Tue 17 Nov
cs.LGstat.AP
pdf html
- reading online learning, manufacturing, online, large
- your work near “Timing as an Action: Learning When to Observe and Act” 0.71
- favorites near “Dynamically Optimal Treatment Allocation” 0.76
Abstract
Predictive maintenance in high-volume manufacturing often faces a domain-incremental learning problem: as new product configurations enter production, prediction models must learn from newly available data while preserving performance on earlier configurations that remain in operation. In this paper, we empirically study the operational value of continual machine learning methods for addressing this challenge in predictive maintenance, using real-world maintenance data from Bosch. The setting at Bosch covers 25 product configurations over five years, which creates a practical challenge for predictive maintenance models, as such models must be updated with data from newly introduced configurations without reducing performance on configurations learned earlier, a phenomenon known as "forgetting". At the same time, retaining all historical sensor data and repeatedly retraining models is often infeasible in high-volume manufacturing since industrial machines generate too large data volumes. To address this challenge, we combine continual machine learning with robust optimization to support predictive maintenance. We then compare continual machine learning with a broad set of baselines, including fine-tuning, domain-adaptation methods, online learning methods, and configuration-specific models. Across experiments, continual machine learning methods achieve stronger predictive performance and lower maintenance costs than the baselines, with a reduction of cost-related indicators of more than 40 % relative to the baselines. Our results demonstrate the operational value of continual machine learning for decision support in maintenance when product portfolios grow over time and historical manufacturing data cannot be fully retained.
John E. Darges
Fri 13 Nov
cs.LGstat.CO
pdf html
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.78
- interest Semiparametric inference and debiased machine learning 0.80
- reading expansions, fourier, estimates, surrogates
Abstract
Random feature expansions (RFEs) provide a flexible approach to surrogate modeling, but their accuracy depends on the distribution used to sample features. We propose spike-and-slab random feature expansions (SAS-RFE) to learn the feature sampling distribution from sensitivity analysis. The method employs a spike-and-slab distribution that assigns probabilities to the inclusion of individual inputs in randomly sampled Fourier features. These inclusion probabilities are determined by estimates of total Sobol' sensitivity indices, rather than by explicitly selecting important inputs or interaction terms. An information theoretic criterion shows that the total Sobol' indices yield a distribution that approximates the input interaction structure of the model. An iterative algorithm updates the sampling probabilities using sensitivity estimates from successive RFE surrogates. Analytical formulas for the Sobol' indices of random Fourier feature expansions enable these updates without additional evaluations of the target function. Numerical experiments on six benchmark problems, with up to 100 input dimensions, demonstrate improvements in surrogate approximation and sensitivity estimation.
Masatoshi Shoji, Isao Yokota, Carolin Herrmann
Mon 16 Nov
stat.ME
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.81
- reading control, historical, gains, data analysis
- favorites near “Bayesian Semiparametric Model for Sequential Treatment Decisions with Informative Timing” 0.80
- looks like clinical and biomedical studies
Abstract
Randomized controlled trials are often infeasible in rare diseases because of limited sample size, high costs, or lengthy durations. Hybrid control designs address this by combining a concurrent control arm with historical data. However, existing methods such as propensity score matching (PSM) borrow historical data only at the analysis stage, which limits power gains when covariate distributions differ between historical and concurrent data. We propose a stratified block randomization method for hybrid control trials. Within each stratum, more concurrent subjects are allocated to the treatment arm than to the concurrent control arm, and the difference is filled with historical controls so that the treatment and hybrid control arms are balanced 1:1; once the available historical controls in the stratum are used or a pre-specified overall cap is reached, allocation returns to 1:1. Historical controls are incorporated either by sampling or by weighting. We compared both methods with PSM in terms of type I error rate, power, and sample size via Monte Carlo simulations. The sampling method maintained the type I error rates when there was neither heterogeneity in the outcome model nor unmeasured covariates. It also had power comparable to or greater than that of PSM, particularly when historical controls were unevenly distributed across strata. The power of the weighting method increased with the pooled historical sample size, but its type I error rate was inflated when covariates beyond the stratification factors were present or when stratification factors were dichotomized from continuous covariates. The proposed method incorporates historical data at the allocation stage rather than only at the analysis stage, enabling more efficient use of historical data.
Leonardo Nascimento Ferreira, Delane Ribas da Rosa, Renata Negri, Marcos Vinicius Gualberto Barbosa da Silva, Rafaella Resende Andrade, Nicole Costa Resende Ferreira
Wed 18 Nov
stat.AP
pdf html
- reading aggregate, productivity, causes, year
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.63
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.60
Abstract
Context: Rising temperatures and warm tropical nights threaten dairy productivity and resilience across tropical landscapes. Objective: This study characterises multi-scale interactions among thermal regimes, nighttime cooling, metabolic heat load, and crossbred genetics on daily milk yield in Brazilian dairy systems. Methods: A 24-year panel (2001--2024) of 788,528 test-day records from 87,809 Girolando (Holstein $\times$ Gyr) cows across 408 municipalities was linked to weather data. Two-way fixed effects, distributed lag, and piecewise threshold spline models were estimated across yield quartiles and crossbred fractions. Results: While aggregate summaries suggest stable yield up to a daily mean Temperature--Humidity Index (THI) of 76, production-stratified splines show high-yielding cows lose milk starting at THI 68 and accelerating above 76 ($-0.162$ kg animal$^{-1}$ d$^{-1}$ unit$^{-1}$, $p=0.0062$), whereas low-yielding cows remain unaffected ($p=0.81$). Distributed lag models reveal an immediate drop followed by a deeper trough four days later. Lack of nighttime relief ($\mathrm{THI}_{\min} \ge 68$) reduces yield by an extra 2.37 kg animal$^{-1}$ d$^{-1}$ ($p \le 0.001$) and causes an 8.6\% loss after seven warm nights, whereas wide diurnal amplitude buffers output by $+1.60$ kg animal$^{-1}$ d$^{-1}$. At comparable high yields ($\approx 31$ kg animal$^{-1}$ d$^{-1}$), intermediate crossbreds ($1/2$ and $5/8$ Holstein) sustain yield up to THI 76, whereas $7/8$ Holstein cows decline at THI 68. Conclusions: Tropical dairy adaptation requires continuous nighttime cooling for high-yielding cows combined with balanced Zebu--taurine crossbreeding.
Karthik C. S., Dor Minzer
Fri 13 Nov
cs.CGcs.CCcs.DSstat.ML
pdf html
- reading la, goal, observe, cover
- favorites near “Random Walks on Simplicial Complexes and Harmonics” 0.77
- your work near “Computational Methods for Economic Models with Function Valued States” 0.71
Abstract
In the Euclidean $k$-means problem, we are given a set of $n$ points in $\mathbb{R}^d$, and the goal is to find $k$ representative points in space so as to minimize the sum of the squared distances from each point to its closest representative. Algorithmically, de la Vega, Karpinski, Kenyon, and Rabani [STOC'03] and Kumar, Sabharwal, and Sen [JACM'10] obtain a $(1+\varepsilon)$-approximation for high-dimensional Euclidean $k$-means in time $2^{(k/\varepsilon)^{O(1)}} \cdot dn^{O(1)}$, which was subsequently improved to $2^{\tilde{O}(k/\varepsilon)} + O(ndk)$ by Feldman, Monemizadeh, and Sohler [SoCG'07]. On the lower-bound side, a recent work of Cohen-Addad, Karthik, Saulpic, and Schwiegelshohn [SoCG'26] proposed the Exponential Time for Expanders Hypothesis (XXH), which asserts exponential-time hardness for $(1+\varepsilon)$-approximating Vertex Cover on vertex expanders, and used it to prove nearly matching running time lower bound whenever $k=\tilde{\omega}(1/\varepsilon)$. We observe that XXH does not hold in the parameter regime needed for the above result, and show how to repair their reduction under a more standard assumption. More precisely, we show that assuming the Almost Exponential Time Hypothesis for 3-SAT (Almost ETH), for every constant $\beta\in(0,1)$, Euclidean $k$-means in $\mathbb{R}^d$ has no $(1+\varepsilon)$-approximation algorithm running in time $2^{(k/\varepsilon)^{1-\beta}}\text{poly}(n,d)$, whenever $k=\tilde{\omega}(1/\varepsilon)$. By modifying the construction, we also obtain a similar result for approximating the Euclidean $k$-median problem.
Irene Boruah
Fri 13 Nov
stat.MEcs.LGstat.AP
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.76
- reading materials, successful, apparent, actually
- interest Empirical Bayes, shrinkage and decision theory 0.78
Abstract
Uncertainty quantification for materials property prediction increasingly reports conditional coverage, coverage within strata of the predicted value, alongside marginal coverage. The worst-bin statistics used for this are downward-biased in finite samples, a known defect that applied materials work does not generally correct for. We replicate a single-dataset conditional coverage study across five regression datasets (n = 437 to 21,263), seven interval methods and 50 random splits each, correcting the worst-bin statistic against its expected value under exact conditional validity. The null is simulated from realized per-bin test counts and referenced to each method's own marginal coverage, separating marginal miscalibration from uneven coverage. On the smallest dataset the correction runs from 0.061 to 0.096, between a third and two thirds of the shortfalls actually reported, against 0.009 to 0.011 on the two largest. Split conformal prediction's failure survives the correction on four of five datasets; two conclusions of the antecedent study do not. Conformalized quantile regression, characterized there as a negative case, fails on one dataset only. Cross-fitted normalized conformal prediction, grouped there with the successful methods, fails on three of five. Group-conditional (Mondrian) conformal prediction stays below our significance criterion on every dataset, yet exceeds the null on three of five under an octile refinement of the variable it calibrates on. A level-matched ablation locates that uniformity in group-conditional calibration rather than in the elevated quantile level, and removes Mondrian's apparent width cost entirely. Conditional coverage evaluations should report the null distribution of the worst-bin statistic used, the method's realized marginal coverage, and the resolution of the evaluation partition relative to any partition the method calibrates on.
Md Rafid Islam, Rafsan Jany, Rafid Bin Zaman, Mohammad Fahim Islam, Hafiz Abdur Rahman
Thu 19 Nov
cs.LGstat.AP
pdf html
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.71
- reading forecasting, nominal, reasonably, large
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.70
Abstract
Next-day air quality forecasting often assumes city records are equally complete and independent. However, an audit of a large Bangladesh dataset revealed a sharp coverage expansion after August 2022 and showed that 103 city labels correspond to only 61 unique AQI trajectories. To address these issues, this paper presents CAFE-AQI, a coverage-aware framework for forecasting next-day maximum AQI. It also estimates the probability of Unhealthy-or-worse conditions (AQI >= 151). Data were divided chronologically into training (2022-2023), calibration (2024), and testing (2025). To prevent duplicate records from dominating training and evaluation, identical trajectories were weighted. The framework combines temporal and pollutant features with a five-neighbor spatial context. It also includes a class-balanced warning head and risk-conditional conformal calibration aimed at providing early warnings and prediction intervals. Evaluation on the duplicate-balanced 2025 data gave an MAE of 9.76, an RMSE of 15.04, and an R-squared value of 0.910 for CAFE-AQI. Its MAE was 3.63 AQI lower than persistence (95% CI: 3.49-3.76), with lower errors in each of the 61 trajectory groups. For unhealthy-air warnings, the F1-score was 0.908 and the AUROC was 0.982. The nominal 90% intervals covered 92.09% of observations. When an entire spatial block was excluded from training, the MAE remained 10.66, showing that the lightweight model transferred reasonably well to unseen locations.
Diego C. Nascimento, Ana C. S. Batista, Francisco Louzada, Paulo H. Ferreira
Thu 19 Nov
stat.APstat.OT
pdf html
- favorites near “Applied Koopmanism” 0.83
- interest Adaptive data collection and anytime-valid inference 0.80
- your work near “Computational Methods for Economic Models with Function Valued States” 0.73
Abstract
The main challenge in monitoring multiple sensors is the high dimensionality that a process can easily reach. From a decision-making perspective, converting these multivariate dynamics into an informative visualization is also not trivial. Statistical Process Control (SPC) techniques detect changes in quality characteristics, but they are difficult to apply beyond the normality (Gaussian) assumption. For instance, the most well-known SPC multivariate control chart is based on Hotelling's $T^2$ statistic, which assumes that the process data follow a multivariate normal distribution (which rarely occurs in practice) and that the observations are serially independent. This study focuses on the multivariate case (dimension $d \geq 2$), regardless of the data dependence or marginal asymmetry. We propose a new control chart for change detection in copula-based multivariate processes, called the $C_n$-chart, which is based on a density level-set tolerance region built from a copula function, and we derive its run-length properties. The methodology also quantifies the (not necessarily linear) dependence among the variables. Two case studies are presented, one on water quality control and the other on monitoring brass-steel thermostats, complemented by a simulation study of run-length performance and computational cost. The results show that the proposed methodology captures both the association among variables and the anomalies, and that it accommodates variable selection through conditional (regression) modeling of the margins.
Ardavan Yazdanbakhsh
Mon 16 Nov
stat.AP
pdf html
- reading did, forecasting, higher, national
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.70
- your work near “Computational Methods for Economic Models with Function Valued States” 0.61
Abstract
Background. Human activities in areas with burnable land cover can lead to wildfire ignitions, and higher levels of human presence increase the likelihood of such fires. Building locations provide a widely available representation of human presence. Aims. This study examines how accurately proximity to buildings predicts wildfire ignition cause across different land-cover types. Methods. Supervised classification methods were applied to building, land-cover, and ignition data for the contiguous United States to estimate the conditional probability that an ignition is human-caused. Proximity to buildings was represented using either building density within surrounding zones or distance to nearby areas with high building density. Key results. Both methods, particularly the distance-based approach, showed that proximity to buildings is a strong predictor of ignition cause: approximately three quarters of both human- and lightning-caused ignitions were correctly classified. Retraining the models separately for individual states did not substantially improve classification performance relative to the national model. Conclusions. Building proximity is an effective predictor of wildfire ignition cause, and the relationship between ignition cause and building-related predictors is similar across many regions of the contiguous United States. Implications. Building-proximity variables can be incorporated into wildfire forecasting and risk assessment models to improve prediction of human-caused wildfire ignitions.
Haoran Yan, Rongjie Lai, Wenjing Liao
Tue 17 Nov
stat.MLcs.LG
pdf html
- interest Kernels, operators and functional data 0.79
- favorites near “Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss” 0.77
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.69
Abstract
Denoising autoencoders have shown strong empirical performance in recovering clean data from their noisy observations, yet their underlying statistical behavior remains underexplored. In this paper, we study denoising autoencoders under the manifold hypothesis, assuming that clean data are supported on low-dimensional smooth manifolds embedded in a high-dimensional ambient space, while noisy observations are generated by isotropic Gaussian noise. We establish generalization guarantees in three geometric scenarios: a single manifold, a pairwise separated finite union of manifolds, and a finite union of intersecting manifolds. Using local chart representations, partitions of unity, and neural network approximation, we derive upper bounds for the expected squared reconstruction error for denoising autoencoder. In the single-manifold and separated-union scenarios, the error consists of a sample-dependent statistical term of order $n^{-2/(d+2)}\log^4 n$, a residual term of order $d\sigma^2/D$ caused by tangential noise, and an exponentially small tail term associated with failure of the manifold projection. Here $D$ is the data ambient dimension, $d$ is the intrinsic dimension of the clean-data manifold, and $\sigma^2$ is the expected squared noise magnitude. In the scenario of intersecting manifolds, we further identify an additional ambiguity term determined by the probability mass near the intersection neighborhood. Numerical experiments on synthetic manifolds and real image datasets support the predicted sample-size scaling law and illustrate the additional reconstruction ambiguity near manifold intersections.
Yichen Hu, Xiucai Ding, Jian Kang
Tue 17 Nov
stat.ME
pdf html
- interest Kernels, operators and functional data 0.80
- favorites near “Operator-valued Kernels for Learning from Functional Response Data” 0.80
- reading eigenfunctions, operator, nonlinear, empirical
Abstract
In this paper, we propose SPASE, an adaptive, sparsity-aware, projection-assisted spectral embedding method for feature extraction from high-dimensional noisy data with underlying nonlinear structures and latent sparsity patterns. We first construct a data-driven projection that adapts to the hidden sparsity patterns and the signal-to-noise ratio, reducing noise-induced distortion of pairwise distances. We then apply kernel-based spectral embedding to the projected data to explore the underlying geometric structure. The key methodological contribution is an energy-based projection selection algorithm that is adaptive to the signal-to-noise ratio and sparsity of the signal representation, while mitigating the impact of noise. On the theoretical side, we show that the resulting embeddings converge to the eigenfunctions of an integral operator associated with the latent signal, and the method can remain effective when the signal-to-noise ratio is low. We demonstrate the empirical performance of the proposed method through simulation studies and real-data applications, showing improved accuracy in downstream tasks such as classification and clustering.
Shuochen Wang, Mark S. Allen, Auroop R. Ganguly
Tue 17 Nov
physics.ao-phstat.AP
pdf html
- reading consistent, gains, basis, ifs
- interest Macroeconometrics and time series 0.73
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.65
Abstract
Accurate tropical cyclone (TC) track guidance is essential for warnings, evacuation planning, and risk communication, but outlying members and distinct track scenarios can degrade raw ensemble forecasts. Statistical and machine-learning (ML) methods proposed to improve ensemble guidance are difficult to compare because previous studies often use different forecast systems, basins, periods, and verification methods. Using a common verification framework, this study compares six objective methods (four statistical trimming and two ML methods) with the full ensemble mean for 138 TCs forecast by GEFS, MOGREPS-G, and IFS during 2016-2024. Dynamic trimming (DT) is the most consistent statistical method, reducing mean total track error by 8.0%, 13.4%, and 16.1% at short, medium, and long ranges. Its largest benefits occur in high-spread cases, where it can reach 70% of the low-error bound. Short-lead-error trimming (SLET) provides useful delayed guidance after early observations become available, but its improvement decreases with lead time. Regression-mixture clustering (RC) shows potential in low-spread cases and provides the largest improvement over the control run. The gradient boosted tree (GBT) produces the largest medium-range improvement and exceeds the theoretical trimming bound, suggesting strong potential, particularly when the observed track is poorly represented by the ensemble family. However, its 12.8% short-range degradation and gains concentrated in a small subset of cases indicate overcorrection when the ensemble mean is already accurate, highlighting the need for additional constraints. These results provide a consistent basis for using objective error-reduction methods to strengthen TC track guidance for operational decision-making.
Peng Sun, Ruoyu Wang, Xue Luo
Tue 17 Nov
stat.COmath.OC
pdf html
- reading kalman, state, nonlinear, variational
- interest Kernels, operators and functional data 0.74
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.70
Abstract
This paper investigates joint state and noise covariance estimation for nonlinear systems subject to dual reliability impairments: random packet dropouts and additive measurement corruption. This scenario poses challenging nonconjugate inference, as clean and corrupted measurements share an unknown baseline covariance $\boldsymbol{R}_k$ with fixed known corruption covariance $\boldsymbol{E}_k$. Packet reception is observable, whereas the authenticity of received measurements remains latent and must be inferred. To resolve this intractable estimation problem, we cast joint posterior inference as a variational optimization problem and develop a conjugate variational inference adaptive Kalman filter with dual reliability (CVIAKF-DR). The proposed method retains analytical updates for the prediction covariance and two reliability probabilities, while adopting CVI to estimate the state and baseline measurement covariance. The covariance update combines the analytical contribution from clean measurements with Monte Carlo gradient estimates for corrupted measurements. Numerical results on a re-entry tracking benchmark demonstrate the superior accuracy of the proposed method in both state and noise covariance estimation. Remarkably, even built upon the linear Kalman filtering (KF) framework that is originally designed for linear systems, our method achieves more accurate estimation performance than dedicated nonlinear filtering algorithms based on the cubature Kalman filter (CKF).
Eyal Weiss
Mon 16 Nov
cs.LGstat.ML
pdf html
- reading large, test error, library, considers
- interest Kernels, operators and functional data 0.74
- favorites near “Why Does Deep Learning Work? - A Perspective From Group Theory” 0.76
Abstract
e-means (Bessa et al., 2026) is a recent mini-batch k-means algorithm, now part of the FAISS library, that reaches the error of Lloyd's algorithm about an order of magnitude faster on large embedding datasets. Every five mini-batches, it removes every cluster whose running point count falls below a tenth of the average and reuses its centre to split a large cluster. In the authors' ablation, removing this mechanism, called dead-cluster revival, raises the error more than removing any other component added to the moving-average update. The rule, however, considers only cluster size: it cannot distinguish a cluster that is small because it is redundant from one that covers a rare, isolated group of points, and on data with such groups it repeatedly removes centres whose removal is costly. We add a second condition, based on the distance between centres: a small cluster is removed only if its centre is close to another centre, relative to a running estimate of the mean squared error. When a centre equals the mean of its points, this squared distance is an upper bound on the average increase in squared error, per point of the cluster, caused by removing the centre. The condition costs one nearest-centre search for the small clusters every five mini-batches and adds no measurable running time. On the e-means paper's embedding benchmarks, it never prevents a removal, so the results are identical to those of FAISS. On datasets with small isolated groups not used during development, it reduces the test error after two epochs by roughly 10% to 80% relative to count-based revival, and it preserves the advantage of k-means++ initialization; on the other evaluation datasets, it stays within about 1% of the better of revival enabled and disabled. We also report two failure modes that the analysis predicts. This is a CPU proof of concept; GPU implementations are left to future work.
Georgios Giannarakis, Vasileios Sitokonstantinou, Dimitrios Bormpoudakis, Ilias Tsoumas, Nikiforos Samarinas, Gustau Camps-Valls, et al.
Fri 13 Nov
cs.LGstat.ML
pdf html
- reading gains, climate, higher, causal
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.60
- your work near “Computational Methods for Economic Models with Function Valued States” 0.51
Abstract
Sustainable agricultural intensification requires a precise understanding of how land use decisions influence soil organic carbon (SOC) sequestration under varying environmental conditions. We apply causal machine learning to evaluate the heterogeneous impact of winter wheat-based rotations on SOC at the field-level across Lithuania. Using a nationwide dataset collected over five years (2018-2022), including crop types, management practices, soil properties, and environmental conditions, we estimate how the effects of winter wheat-based rotation vary across space and climate. Our results indicate that winter wheat generally enhances SOC levels, but its effectiveness is modulated by climate. Cooler regions experience higher SOC gains, whereas warmer regions show diminished or even negative effects. Future climate projections suggest that the potential for winter wheat carbon sequestration may decline under high-emission scenarios. These findings highlight the importance of localized agricultural strategies that account for climate variability to maximize SOC sequestration and ensure long-term soil sustainability.
Samuel Pinilla
Tue 17 Nov
math.OCeess.SPmath.MPstat.AP
pdf html
- reading reflections, semidefinite, band-limited, frequencies
- interest Kernels, operators and functional data 0.68
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.63
Abstract
The ambiguity function (AF) of a radar waveform describes its ability to resolve targets in range and Doppler. Designing a waveform that attains a prescribed AF magnitude is a phase retrieval problem with four trivial ambiguities, and a time- or band-limited pulse is determined up to these ambiguities by at most three times as many AF samples as it has time samples or frequencies. Existing solvers for the narrowband AF, however, are non-convex: they cannot guarantee that they reach the global minimizer, which is unique only up to the trivial ambiguities, so the recovered pulse depends on the starting point. This letter addresses this gap by lifting the problem to positive semidefinite matrices supported on the pulse band, where each AF sample becomes the magnitude of a linear functional. In the lifted space, recovery is formulated as a convex program, and a dual certificate provides a sufficient condition for exactness. The proposed algorithm, wavex, combines an initialization by relaxed reflections, convex semidefinite surrogate steps, and a local polish. Numerical experiments show that wavex accurately recovers band-limited pulses from complete AFs and from AFs with most of their samples removed, provided the remaining samples determine the pulse, and that its accuracy degrades gracefully as the noise increases. wavex is built on the open-source library invexapi, and the code that reproduces every result is available at https://github.com/samuelpinilla/waveX.
M. Hadi Sepanj, Benyamin Ghojogh, Saed Moradi, Paul Fieguth
Wed 18 Nov
cs.CVcs.LGstat.ML
pdf html
- favorites near “Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss” 0.82
- interest Kernels, operators and functional data 0.73
- reading view, empirical, implicit, top-1
Abstract
Self-Supervised Learning (SSL) aims to align augmented views while preserving representation diversity and preventing collapse. Existing non-contrastive methods often address these goals using multiple weighted loss terms. We introduce Centroid Spectral Coupling (CSC), a single-term SSL objective that implicitly couples view alignment, representation diversity, and collapse resistance. CSC constructs centroids from independently normalized augmented-view pairs and applies a ridge-based spectral objective to their uncentered second-moment matrix. Preserving centroid magnitudes encodes view agreement while encouraging balanced representation energy across spectral directions. We establish an exact decomposition into nonnegative alignment and spectral-imbalance components, characterize the geometry of global optima, and derive finite-loss guarantees for alignment and diversity. We also show that complete constant-output collapse is a strict saddle in normalized representation space. Experiments on CIFAR-10, CIFAR-100, and Imagenette show that CSC achieves the highest linear Top-1 accuracy among the evaluated non-contrastive methods on all three benchmarks, and the highest kNN Top-1 accuracy on both CIFAR datasets. A ridge-parameter sensitivity study reveals an empirical alignment--diversity trade-off: increasing $\kappa$ strengthens view alignment but reduces representation effective rank. These results show that a unified spectral objective yields competitive visual representations without explicit contrastive negatives or separately weighted alignment and diversity penalties.
Fernando A. Otero
Thu 19 Nov
stat.ME
pdf html
- reading state, operator, special, relate
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.68
- interest Kernels, operators and functional data 0.69
Abstract
Neutrosophic annotation frameworks for quantum measurement attach an external triple $(t,i,f)$ to each Born probability, and explicitly leave open how indeterminacy could be defined operationally. We show that indeterminacy cannot be recovered from outcome probabilities, but can be computed from the pair (state, effect). For a binary POVM $\{A, I-A\}$ we define the instrument indeterminacy $\mathcal{I}_\rho(A) = 4 \mathrm{Tr}(\rho(A-A^2))$, which vanishes exactly when the state only sees the sharp part of the effect. The normalized Gini index of the outcome splits as $4TF = \mathcal{I}_\rho(A) + 4 \mathrm{Var}_\rho(A)$, a law of total variance that separates the instrument contribution from the state contribution; the identity is a special case of Massar's POVM uncertainty operator, and our contribution is its neutrosophic reading. Moving the system-apparatus cut to a Naimark dilation transfers exactly $\mathcal{I}_\rho(A)$ from indeterminacy to variance, which yields the missing operational definition. We extend the decomposition to $m$-outcome POVMs, relate the minimum indeterminacy of two incompatible unbiased qubit observables to their commutator through Busch's joint-measurability criterion, and show that $\mathcal{I}_\rho(A)$ equals twice the disagreement probability of a repeated L\"uders measurement. The resulting estimator needs neither the state nor the effect and, in simulation, requires 2.5 to 6 times fewer shots than detector tomography for the same error.
Benjamin Remy, Raphael Rossellini, Chihway Chang, Rebecca Willett
Fri 13 Nov
cs.LGstat.MEstat.ML
pdf html
- reading little, choice, inverse, seek
- favorites near “Monte Carlo Geometry Processing” 0.77
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.69
Abstract
Conditional generative models have become the tool of choice for sampling posteriors in scientific inverse problems, where we seek to sample images or signals conditioned on a set of indirect and noisy measurements. Yet they are often overconfident, which is a critical failure in scientific applications where accurate uncertainty quantification is essential. Existing regularizers fall short in complementary ways: differentiable coverage objectives require the model's density and do not identify the posterior, while strictly proper scoring rules such as the energy score need only samples but do not directly enforce coverage. We close this gap by introducing a differentiable loss function based on the expected coverage probability, estimated from samples alone. It applies to any conditional generative model, with no need for tractable densities or architectural changes. We prove that it is a strict divergence, vanishing if and only if the model equals the true posterior. On simulation-based inference benchmarks, it matches a density-based coverage loss without evaluating the density. On high-dimensional imaging inverse problems, it yields better-calibrated posteriors than scoring-rule regularizers, with little change in the reconstruction quality of point estimates.
Jinrui David, Yun Peng
Thu 19 Nov
stat.MLcs.AIcs.LG
pdf html
- your work near “Timing as an Action: Learning When to Observe and Act” 0.74
- interest Computational methods for heterogeneous agent and dynamic economic models 0.72
- reading estimates, interventions, absolute, evidence
- looks like clinical and biomedical studies
Abstract
Predicting the transcriptional response to combined perturbations requires distinguishing interactions between interventions from errors in their individual effect estimates. The latter become more pronounced when single-perturbation evidence is sparse or withheld. We introduce a conditional flow-matching model that estimates the reliability of single-gene effects from training-only cross-fitting and uses those estimates to calibrate the additive source distribution. The velocity is decomposed into a known correction of the additive reference and a learned residual transport field. On Norman genetic perturbations, reduces mean absolute error (MAE) from 0.142 to 0.131 for unseen combinations of observed singles and from 0.198 to 0.178 when constituent singles are withheld. On held-out ComboSciPlex drug combinations, MAE decreases from 0.191 to 0.176. Non-additive residual correlation improves from 0.574 to 0.638, and component ablations identify the contribution of reliability weighting. The results indicate that accounting for uncertain constituent effects improves generalization to unseen genetic and chemical combinations, with a moderate increase in training cost.
Xiyao Chen, Takumi Sato, Linyu Peng
Tue 17 Nov
eess.SPcs.LGstat.AP
pdf html
- favorites near “Neural Operator: Learning Maps Between Function Spaces With Applications to PDEs” 0.78
- interest Kernels, operators and functional data 0.75
- reading matrices, numerical methods, procedures, positive definite
Abstract
Matrix information geometry (MIG) detectors exploit the Riemannian geometry of Hermitian positive definite (HPD) matrices to achieve robust object detection in nonhomogeneous clutter. The matrix constant false alarm rate (matrix-CFAR) method is a representative MIG detector that relies on the geometric structures with respect to specific metrics. In most cases, computing representative HPD covariance matrices requires numerical methods, typically involving computationally intensive iterative optimization algorithms. Although explicit formulas are available for certain cases, such as the log-Euclidean (LE) mean, they are generally unavailable for many other matrix means and medians. To address this challenge, we propose integrating the Hermitian positive definite neural network (HPDNet) into the matrix-CFAR framework, replacing conventional numerical optimization procedures with a learned, geometry-preserving mapping that directly estimates representative HPD covariance matrices from data. In the simulations, the framework is evaluated for MIG detectors using both means and medians associated with the affine-invariant Riemannian metric, the LE metric, and the Bures--Wasserstein metric, on both simulated compound-Gaussian clutter and real IPIX sea clutter. In both cases, the HPDNet is trained solely on simulated data to assess its applicability to the study of real data. The results show a significant reduction of computational cost in testing, while maintaining detection performance and robustness.
Batuhan Avci, Marko Susnjar, Marko Maljkovic, Nikolas Geroliminis
Fri 13 Nov
stat.APcs.AIcs.LGcs.SYeess.SY
pdf html
- your work near “Differentiable State Space Models and Hamiltonian Monte Carlo Estimation” 0.74
- reading gains, data provide, gaussian process, modern
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.67
Abstract
Real-time urban traffic monitoring is essential for modern city management, yet conventional static sensors provide only limited spatial coverage. Autonomous drone fleets can complement this infrastructure by offering flexible, high-resolution mobile sensing across different parts of the network. This paper introduces a spatio-temporal Gaussian process (GP) framework for reconstructing network-wide traffic states from partial aerial observations. The model accounts for the physical structure of the road network through a graph Mat\'ern kernel, which embeds network topology into the spatial covariance structure, and captures temporal dependencies through an exponential temporal kernel. Within this framework, historical loop detector data provide an informative prior mean that is updated in real time using incoming drone observations, yielding network-wide point estimates together with predictive uncertainty. We validate our framework in two distinct case studies: a microscopic simulation of Barcelona and real-world drone data from Athens. The results demonstrate that our approach outperforms historical baselines in estimating and predicting Vehicle Hours Traveled (VHT), notably reducing the required fleet size. Furthermore, leveraging the model's predictive uncertainty to guide drone movements yields additional performance gains, highlighting the potential for future uncertainty-aware fleet routing algorithms.
Ziqi Mu, Qiuyi Wu, Zhicheng Ji, Pixu Shi
Tue 17 Nov
stat.MEq-bio.QM
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.73
- interest Kernels, operators and functional data 0.75
- favorites near “Hypoelliptic diffusion maps I: tangent bundles” 0.80
Abstract
Simultaneous profiling of multiple molecular layers at spatially resolved locations creates opportunities to identify shared cross-modal biological variation in its tissue context. We present multiCAsp (multi-omic Canonical Association informed by spatial graph), a probabilistic latent-factor model for integrating $M \geq 2$ spatially resolved omic modalities. It incorporates spatial dependence through a graph-Laplacian prior on shared latent factors and uses an expectation-maximization algorithm to estimate factor scores and modality-specific feature loadings. Because the prior is defined on a neighborhood graph rather than on Euclidean distances, it accommodates irregular tissue geometries and extends naturally to multiple tissue sections. Simulations show that multiCAsp consistently achieves higher mean latent subspace and loading recovery than alternative methods across varying sample sizes, feature dimensions, and noise levels, with the largest gains under high noise. Compared with an existing Gaussian process-based spatial factor model, it is also less sensitive to how the spatial coordinates are represented, produces spatially smoother factors, and has lower computational cost. Applied to metabolomic-proteomic data from macaque cortex and gene-protein data from human renal carcinoma, multiCAsp recovers spatially coherent multi-omic factors with interpretable biological signatures. In the macaque data, transferred factors retain broad spatial patterns across animals and support cross-modal prediction of protein abundances from metabolomic measurements.
Marie Kempkes, Jakob Spiegelberg, Evert van Nieuwenburg, Vedran Dunjko
Fri 13 Nov
cs.LGstat.ML
pdf html
- reading theory, little, procedures, large
- interest Online learning and statistical learning theory 0.78
- your work near “Timing as an Action: Learning When to Observe and Act” 0.70
Abstract
A large body of work in quantum machine learning has centered on characterizing the expressivity and performance of quantum models. In contrast, little attention has been paid to the reliability of predictions beyond average performance measures such as accuracy. In this work, we introduce an uncertainty-aware quantum classification framework based on Dempster-Shafer theory. This theory allows to express uncertainty through weights assigned not only to individual classification labels but to sets of labels. We introduce a training strategy that enables parameterized quantum circuits to learn such assignments directly from data. This provides an efficient way to prepare the corresponding quantum states and enables the application of different post-processing strategies. In particular, it offers a practical setting for quantum Dempster-Shafer fusion procedures with polynomial resource requirements, even when the corresponding mass functions are defined over exponentially many subsets. Overall, our work represents a step toward enabling quantum models to express uncertainty about their predictions, thereby extending the evaluation of quantum models beyond predictive performance alone.
Ghurumuruhan Ganesan
Tue 17 Nov
stat.MLcs.ITcs.LGmath.IT
pdf html
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.75
- favorites near “Hypoelliptic diffusion maps I: tangent bundles” 0.79
- interest Kernels, operators and functional data 0.72
Abstract
In this paper, we study reduced similarity decompositions of \emph{deterministic} datasets using redundancy graphs and obtain \emph{ensemble} bounds for the minimum size of such constrained decomposition. We consider batch decompositions that reduce similarity \emph{across} batches and use the Lov\'asz local lemma to obtain bounds for the minimum size of a decomposition with a given pairwise dissimilarity. We then consider unique batch decomposition with a given relaxation parameter and obtain upper and lower ensemble bounds for the minimum size. Throughout, we use Binomial random graphs to obtain the desired ensemble bounds, by conditioning on the number of similar \emph{pairs} of data points.
Sergey S. Kuzin
Thu 19 Nov
stat.MEcs.ITeess.SPmath.IT
pdf html
- reading spectrum, methods practical, harmonic, methods common
- favorites near “Monte Carlo Geometry Processing” 0.75
- interest Kernels, operators and functional data 0.72
Abstract
This paper introduces a unified framework for angular super-resolution in antenna arrays. By integrating linear prediction beyond the physical aperture, entropy variations, and harmonic spatial spectra averaging, we recast classical methods - including Capon, Pisarenko, Maximum Entropy, and MUSIC - into a single structure. We extend this framework to accommodate ESPRIT-type methods by treating the number of simultaneously predicted subarray channels as an additional dimension. Furthermore, we adapt this approach to non-Gaussian signals, establishing a parallel framework for Virtual-ESPRIT-type methods. As a practical application, we propose universal, J-orthogonalization-based algorithms that estimate spatial super- resolution spectra within a unified computational architecture. This architecture produces spatial spectrum estimates for multiple super-resolution methods and any intermediate variations.
Tapu Datta, Hafiz Imtiaz
Fri 13 Nov
cs.LGstat.ML
pdf html
- reading macro, class, seen, receives
- favorites near “Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss” 0.72
- interest Language models as statistical objects 0.67
Abstract
Federated learning over distributed intrusion-detection sensors naturally produces \emph{class-concentrated} non-IID data, where an entire attack class is held by a single client and absent elsewhere. We show this collapses federated averaging in a specific, localized way: the output layer holds one decision boundary per class, and when a class's aggregation weight comes overwhelmingly from clients that have never seen it, its neuron receives an active anti-signal rather than mere dilution, collapsing to zero. On CIC-DDoS2019, standard FedAvg drives the concentrated LDAP class to $0.00 \pm 0.00\%$ F1 across every seed. We propose Class-Competence Aggregation (CCA), which aggregates the output layer per class using each client's own locally reported validation F1 as the weight, while hidden layers remain under standard size-weighted FedAvg. CCA requires no server-held auxiliary data and adds only one F1 scalar per class per round beyond the feature-scaling and class-count exchanges common to any federated pipeline. On our primary CNN-LSTM architecture, CCA alone raises LDAP F1 from $0\%$ to $84.83 \pm 1.65\%$, and the full method reaches $91.64 \pm 0.57\%$ macro F1, closing the federated-to-centralized gap from $-42.53$ to $-4.05$ points. Benchmarked against four other federated methods under identical partitions and seeds, CCA is the only one that recovers the concentrated class substantially and reliably; a $178\times$ smaller MLP shows the same qualitative recovery, though less completely, confirming the effect holds across architecture scale even as its magnitude varies. Results are validated across three seeds and replicated on a second dataset, CICIDS2017.
Xinyi Hong, Hai Shu, Zhaoyang Zhang, Lei Chen, Yanfeng Yang, Ziqi Chen
Tue 17 Nov
cs.LGstat.ML
pdf html
- reading economic, considerable, message passing, absent
- favorites near “Recent Developments in Machine Learning Methods for Stochastic Control and Games” 0.71
- interest Language models as statistical objects 0.66
Abstract
During the pipeline transportation of refined oil products, the inevitable formation of mixed oil segments poses considerable economic challenges to pipeline networks. Therefore, it is important to accurately predict the length of these mixed oil segments. Although previous methods have achieved relatively accurate predictions by integrating machine and deep learning algorithms, they ignore the statistical dependencies that remain among real-world mixed oil samples, which degrades model accuracy. To address this issue, we propose a novel Sample Dependencies Incorporated Graph Representation Learning (SDI-GRL) model. By representing samples and pipeline features as nodes of a graph, SDI-GRL connects dependent samples through their shared features, allowing message passing to propagate information across related samples. This graph-based framework allows our model to capture the statistical dependencies among samples absent in prior works. A case study on a mixed oil dataset demonstrates that our SDI-GRL model surpasses the strongest existing approach, reducing both the RMSE and the MAE by more than 10%.
Yuri Taglieri Sáo, Geraldo de Freitas Maciel, Julian Cardoso Eleutério
Thu 19 Nov
stat.APphysics.flu-dynphysics.geo-ph
pdf html
- reading dimensionality, possible, presents, control
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.67
- your work near “Computational Methods for Economic Models with Function Valued States” 0.65
Abstract
On 25th January 2019, Dam B1 in Brumadinho, Brazil, suddenly collapsed and produced a catastrophic tailings flow with severe human, environmental and infrastructural consequences. Previous numerical back-analyses of the tailings flow have relied on deterministic approaches, which provide useful reference results but do not explicitly incorporate uncertainty associated to rupture hydrograph and tailings' physical properties. This work presents a probabilistic approach to study the tailings dam-breach flow caused by the rupture of Dam B1, focusing on the quantification and interpretation of the effects of input uncertainty on hazard-related hydrodynamical outputs. Deterministic data was obtained through HEC-RAS v6.6 by considering tailings as an one-phase Bingham fluid. The probabilistic framework combines dimensionality reduction, metamodeling and flooding classification to perform fast prediction of unexplored input scenarios. It allows to obtain data distributions that reflect output uncertainty and to estimate variance-based sensitivity indices to quantify the influence of input uncertainty over output variability. Results show that uncertainty is strongly dependent on time and spatial location. Hydrograph-related parameters mainly control the timing and magnitude of the early flow, whereas yield stress becomes more influential in regions associated with flow attenuation, spreading and possible arrest. Overall, the framework provides probabilistic hazard estimates and insights into the physical mechanisms governing the dynamics of tailings flow.
Yijin Zeng
Wed 18 Nov
cs.CVstat.ML
pdf html
- favorites near “Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss” 0.78
- reading just, importantly, data-efficient, bayesian posterior
- interest Kernels, operators and functional data 0.71
Abstract
Contrastive Language-Image Pre-training (CLIP) achieves impressive zero-shot image classification performance by comparing image embeddings against class-specific text embeddings. While recent few-shot methods leverage labeled training samples to improve upon this zero-shot baseline, they often predominantly rely on image embeddings and underutilize the semantic structure of the text embeddings. In this paper, we introduce DEFTS, a data-efficient, training-free few-shot classification method strictly in the text subspace. We first demonstrate empirically that while raw text embeddings can be suboptimal as direct classifiers, the linear subspace they span remains highly discriminative. DEFTS therefore proposes to reduce the image-text modality gap by projecting estimated image prototypes directly into this subspace. Importantly, by expressing these prototypes as linear combinations of text embeddings, DEFTS reduces a high-dimensional estimation problem to the derivation of just a few coefficients. These coefficients are then computed via a Bayesian posterior mean, using a prior naturally centered on the original text embedding coordinates. Extensive experiments across 12 image classification benchmarks demonstrate that DEFTS significantly improves upon existing training-free methods.
Tao Lian, Antonio M. López
Thu 19 Nov
cs.AIstat.ML
pdf html
- interest Language models as statistical objects 0.73
- favorites near “Inference with Hamiltonian Sequential Monte Carlo Simulators” 0.72
- reading closed-loop, distill, higher, best
Abstract
Federated learning enables collaborative model training without sharing datasets. However, existing federated learning methods for autonomous driving typically require multi-round communication and lack closed-loop evaluation, while one-shot methods mainly focus on perception tasks. To address these limitations, we propose \method, a one-shot federated learning method for end-to-end autonomous driving. It comprises multi-client calibration, anchor-based feature alignment, and token-wise consensus distillation to distill knowledge from heterogeneous models. Closed-loop evaluation on Bench2Drive demonstrates the effectiveness of \method in both one-shot and multi-round federated learning. In the one-shot setting, it achieves a Driving Score of 55.85, which is 4.07 points higher than prior work. In the multi-round setting, it reaches 60.35 after five communication rounds, the best performance among all compared methods under the same number of communication rounds.
Wen Zhang, Kaiyi Pang, Minhao Bai, Shuyu Gan, Liping Zhu
Thu 19 Nov
cs.CRstat.AP
pdf html
- your work near “Perturbation Methods for Incomplete Markets Economies” 0.74
- interest Kernels, operators and functional data 0.73
- favorites near “The Consensus Game: Language Model Generation via Equilibrium Search” 0.75
Abstract
As LLM-generated content becomes increasingly widespread, reliable attribution calls for multi-bit watermarks that embed recoverable messages without distorting the output distribution. We introduce PIVOT, a multi-bit watermark with model-free verification based on mirrored rank-interval sampling. At each generation step, PIVOT uses a keyed pseudorandom scalar and a secret permutation of the vocabulary. The message bit determines whether next-token probability intervals are arranged in forward or reverse secret-rank order. Sampling from these mirrored layouts preserves every token probability exactly while inducing an opposite-signed dependence between the scalar and the sampled token's secret rank. This dependence yields a pivotal rank statistic whose null distribution is invariant to the unknown next-token law, enabling model-free calibration and decoding. We formulate joint watermark detection and message recovery as a multiple-hypothesis testing problem, derive a moment-robust minimax-optimal detection score over a relaxed moment class, and establish finite-sample recovery guarantees. Extensive experiments on two representative large language models validate the effectiveness of PIVOT, demonstrating reliable watermark detection and message recovery while preserving the generator's output distribution.
Zihan Wang, Hyun Sang Park, Ali Rachidi, David Elbert, Todd C. Hufnagel, Wei Chen
Mon 16 Nov
stat.MLcs.LG
pdf html
- favorites near “Monte Carlo Geometry Processing” 0.76
- your work near “Timing as an Action: Learning When to Observe and Act” 0.70
- interest Kernels, operators and functional data 0.70
Abstract
Autonomous experimentation systems require characterization models that can interpret heterogeneous measurements and maintain predictive performance as data accumulate. Developing such models involves addressing domain mismatch between data sources, overconfident predictions on unfamiliar inputs, and catastrophic forgetting during sequential updates. In this work, we present a proof-of-concept framework for uncertainty-aware continual interpretation of two-dimensional X-ray diffraction (XRD) data. The framework integrates established components: maximum mean discrepancy-based domain adaptation to reduce simulation-to-experiment discrepancies, a convolutional neural network with a spectral-normalized neural Gaussian process (CNN--SNGP) for predictive uncertainty estimation, and replay with knowledge distillation for sequential model updating. We evaluate the framework using simulated and previously-collected experimental data for crystal structure classification of single-phase body-centered cubic, face-centered cubic, and hexagonal close-packed patterns. Data are organized into predefined sequential blocks to assess adaptation and retention of previously learned knowledge. Within this setting, the results show accurate classification across simulated and experimental data, improved alignment between the two domains, and reduced forgetting compared with na\"ive fine-tuning. Predictive uncertainty is also evaluated on unfamiliar and noisy inputs. The contribution is the integration and evaluation of these capabilities for sequential XRD interpretation. The framework provides a basis for future integration into autonomous experimentation systems, with experimental planning, instrument control, and closed-loop deployment remaining areas for future development.
Giuseppe Nicolosi, Michael Bonkowski, Carlo Febbi, Christian Mulder
Tue 17 Nov
q-bio.PEq-bio.QMstat.AP
pdf html
- reading differentiation, little, functional, processes
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.60
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.59
Abstract
1. Soil nematodes occupy multiple trophic positions and contribute to key processes in soil food webs. Their communities vary across land uses and environmental conditions, yet it remains unclear whether these differences mainly reflect taxon replacement or nested patterns of taxon occurrence. 2. Here, we analysed two independent European datasets of soil nematode communities, SOB4ES, based on eDNA metabarcoding, and NemEu, based on morphological identification, to examine how genus composition varies across land uses. We partitioned community dissimilarity into turnover and nestedness-resultant components and compared sites under contrasting land uses with those sharing the same land use, while accounting for geographical and soil-related variation. We also assessed matrix-level nestedness, differences among trophic groups and associations between genus composition and environmental gradients. 3. In both datasets, genus turnover was consistently greater between contrasting land uses than within the same land use. Although the differences were modest, they remained significant after accounting for geographical distance and genus richness, whereas nestedness-resultant dissimilarity contributed little to land-use differentiation. These findings suggest that land-use-associated differences in European nematode communities are expressed primarily through changes in genus identity rather than nested patterns of taxon occurrence. 4. The consistency of this result across independently assembled datasets supports the broader relevance of taxonomic turnover for understanding soil biodiversity patterns across European land uses.
Aminu Yusuf, Shakeel Muhammad Ibrahim, Sulaimon Oyeniyi Adebayo, Muhammad Sohail Ibrahim, Muhammad Shahid Jabbar, Shujaat Khan
Fri 13 Nov
cs.CVeess.IVeess.SPstat.ML
pdf html
- interest Language models as statistical objects 0.74
- favorites near “Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss” 0.75
- your work near “Computational Methods for Economic Models with Function Valued States” 0.67
Abstract
Face presentation attack detection requires representations that distinguish physical spoofing cues across changes in acquisition devices, subjects, and attack media. Large visual encoders offer transferable representations, but conventional binary fine-tuning supervises their final decisions without explicitly constraining how local facial evidence is represented under partial observation. We present CP-JEPA, a context-predictive adaptation framework that combines low-rank tuning of a pretrained CLIP image encoder with a joint-embedding predictive objective. A student predicts complete-image teacher features at masked spatial locations, while unmasked classification, masked classification, and prediction consistency retain the connection to the live/spoof task. An exponential-moving-average teacher supplies evolving latent targets. The auxiliary components are used only during training; inference requires only the LoRA-adapted CLIP visual encoder and binary classifier. {We evaluate CP-JEPA under single-source and multi-source OCIM transfer, additional CelebA-Spoof training, transfer to CelebA-Spoof, two-source training, and unseen-attack evaluation on SiW-Mv2.} On the standard OCIM benchmark, CP-JEPA achieves 3.74\% mean HTER and 98.73\% mean AUC. {Single-source evaluation achieves 11.36\% mean HTER across twelve source--target combinations.} With CelebA-Spoof added to the three OCIM training sources, CP-JEPA reduces mean HTER from 3.45\% to 2.35\% and increases mean AUC from 98.68\% to 99.32\%. On SiW-Mv2, it achieves 3.36\% mean HTER across the held-out attack categories. Controlled loss ablations examine the contributions of latent prediction, masked classification, and prediction consistency. The results show that context-predictive adaptation can improve cross-dataset PAD performance, although the gains vary with the training sources and evaluation setting.
Nuria Torrado
Mon 16 Nov
math.STmath.PRstat.MEstat.TH
pdf html
- your work near “Perturbation Methods for Incomplete Markets Economies” 0.72
- interest Semiparametric inference and debiased machine learning 0.77
- favorites near “Applied Koopmanism” 0.77
Abstract
Understanding how dependence among component lifetimes affects the reliability of redundant systems is a central problem in reliability engineering. In this paper, we develop a framework for quantifying the impact of multivariate dependence on $k$-out-of-$n$ system reliability relative to the independent benchmark. Using an orthogonal Legendre expansion of the copula density, we derive an exact decomposition that separates the dependence structure from the system architecture. This representation identifies how individual dependence contributions are transmitted to system reliability through explicit structural weights. An application to multivariate Farlie-Gumbel-Morgenstern copulas demonstrates that higher-order dependence can produce distinct reliability effects even when all bivariate copulas remain unchanged. Our findings provide a systematic approach to quantifying and interpreting the effects of multivariate dependence on the reliability of redundant systems.
Mansoure Giahi Sabour, Hassan Yazdanian, Babak Maboudi Afkham
Fri 13 Nov
stat.APcs.NAmath.NAmath.PR
pdf html
- reading sequential, connection, computed, presents
- favorites near “Monte Carlo Geometry Processing” 0.75
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.68
Abstract
Sparse-angle X-ray computed tomography (CT) presents a challenge for reliable inference, as limited measurements may leave object boundaries unresolved. Bayesian optimal experimental design (BOED) methods select informative measurements, but can be computationally expensive or require data-driven surrogates and do not explicitly exploit microlocal visibility. We address this limitation by focusing on estimating inclusion boundaries rather than the entire attenuation field. For piecewise-constant objects, these boundaries correspond to singularities characterized by their wavefront sets. Microlocal analysis determines the X-ray view-angles at which these singularities are visible. We combine this information with Bayesian uncertainty quantification to introduce B-MUSE, a microlocal uncertainty-driven sequential experimental design method. We define a microlocal C-optimality criterion associated with the wavefront direction of the most uncertain boundary singularity and prove that its optimal solution satisfies the microlocal visibility condition. B-MUSE uses this result to sequentially select informative view-angles according to boundary uncertainty. It also incorporates an adaptive stopping criterion that terminates acquisition when additional measurements provide no meaningful uncertainty reduction. We evaluate B-MUSE through numerical experiments, including interpretable examples, prior-generated inclusions, and a multi-inclusion example from an external dataset, and benchmark it against recent methods. The results show that B-MUSE substantially reduces posterior uncertainty with few adaptively selected measurements while providing boundary estimates. These findings establish a direct connection between BOED and microlocal visibility and demonstrate the potential of uncertainty-driven acquisition to reduce measurement requirements for goal-oriented X-ray CT without training data.
Harry O'Donnell, Antony M. Overstall
Fri 13 Nov
stat.AP
pdf html
- your work near “Estimating Distributional Responses to Macroeconomic Shocks” 0.67
- interest Kernels, operators and functional data 0.72
- reading gmm, tactical, modern, assign
Abstract
The increasing availability of detailed football performance data has created new opportunities for identifying player archetypes and informing recruitment and tactical decision-making. Most existing clustering approaches assign each player to a single group, despite the reality that many modern footballers perform hybrid roles spanning multiple stylistic categories. Performance data for 314 outfield players from the 2024/25 English Premier League season are analysed using a Gaussian mixture model (GMM) applied to principal component representations of player statistics. Unlike hard-assignment methods such as k-means, the GMM provides probabilistic cluster memberships, allowing uncertainty and role hybridity to be quantified. Seven player archetypes are identified in the full dataset. Posterior membership probabilities reveal a subset of players whose profiles span multiple clusters, providing an objective measure of tactical flexibility and stylistic ambiguity. Further analyses within defender, midfielder, and forward groups recover position-specific sub-archetypes and highlight exceptional outliers. The results demonstrate that model-based clustering provides a flexible and interpretable framework for identifying latent structure in high-dimensional performance data while explicitly quantifying uncertainty in cluster membership. More broadly, the approach is readily transferable to other sports and domains in which individuals exhibit overlapping characteristics and cannot be adequately represented by hard cluster assignments alone.
Juyeon Kim, Chaeyeon Yi, Yun Am Seo
Tue 17 Nov
stat.APcs.LGphysics.ao-ph
pdf html
- reading versus, gains, literature, significant gains
- interest Language models as statistical objects 0.69
- your work near “Timing as an Action: Learning When to Observe and Act” 0.66
Abstract
Frost protection requires local warnings where leaf-sensor inputs are unavailable. In the frost literature we surveyed, physics entered machine-learning models through inputs, numerical-model coupling, or losses, rather than model structure. We develop a transformer-compatible module that embeds leaf energy-balance physics in freezing-probability forecasts: an energy-balance temperature reference and a learned residual predict leaf temperature, which drives a multiplicative gate imposing a temperature-dependent probability cap. We mathematically prove conditional monotonicity and an alarm boundary, and test whether placing the same information in the output pathway or in attention yields different benefits. We evaluated 12 hourly leads at 19 Korean stations over five frost seasons, with and without leaf-sensor inputs. In the 2025-2026 hold-out season without these inputs, our time-token transformer (PET-E) achieved a lead-mean critical success index (CSI) of 0.516 versus 0.478 for fixed-configuration, joint-lead LightGBM, with significant gains at all 12 leads, and the fraction of alarms at observed leaf temperatures of at least 2 degrees C fell from 0.087 to 0.051 relative to the unmodified backbone. Across the hold-out and a retrospective 2024-2025 evaluation, the module reduced Brier scores and warm-surface alarm fractions in both the time-token transformer and iTransformer, with no significant CSI decrease at any lead. At held-out stations without static or leaf-sensor inputs, both module-equipped backbones exceeded LightGBM at all 12 leads (CSI 0.468 and 0.467 versus 0.437). In the time-token backbone, the temperature pathway reduced this alarm fraction in all four settings, whereas the attention prior increased CSI in three. These findings link an explicit temperature-dependent probability cap to improved forecasts and distinguish output-side from attention-based embedding.
Nina Effenberger, Florian E. Roemer, Maybritt Schillinger, Maxim Samarin, Erich Fischer, Reto Knutti
Fri 13 Nov
cs.LGstat.AP
pdf html
- reading climate, boosting, example, solution
- interest Macroeconometrics and time series 0.70
- favorites near “The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding” 0.64
Abstract
Extreme weather events are critical for climate impact assessments. While physical models can simulate unprecedented extremes, for example, through ensemble boosting, this approach is computationally demanding, often restricting output to coarse spatial resolutions. Probabilistic generative downscaling offers a solution by enabling efficient sampling from the entire climate distribution, including its extreme tails. Here, we first demonstrate that machine learning generated regional climate model heatwaves over Paris are physically realistic. We then leverage an emulated machine learning dataset to perform analog-based downscaling of coarse global climate model ensemble boosting events. Our hybrid framework effectively bridges high-resolution regional detail with rare-event sampling, providing an efficient tool for extreme event downscaling and capturing multi-model uncertainty.
Taro Hayashi, Ryoichi Suzuki
Fri 13 Nov
math.NTmath.STstat.TH
pdf html
- your work near “Solution of Rational Expectations Models with Function Valued States” 0.70
- interest Adaptive data collection and anytime-valid inference 0.76
- favorites near “Introduction to regularity structures” 0.75
Abstract
We study exact recovery of the Hasse--Weil zeta numerator from repeated noisy observations of a randomized Euler product. For a genus-two curve over $\mathbb{F}_q$, we attach independent symmetric signs to its closed points and observe the logarithm of the resulting product at $t_q=c/\sqrt q$, where $0<c<1/\sqrt{12}$ is fixed. For every fixed, known additive noise density $f$ satisfying $\sqrt{f}\in H^1(\mathbb{R})$, we prove that the minimax number of independent observations required for exact recovery with error probability at most $\delta$ has order $q^4\log(1/\delta)$. No moment condition is imposed on the noise, so the result includes Cauchy readout. The upper bound estimates the phase of an empirical characteristic function and uses the separation of the integral Weil coefficients. The matching lower bound compares the full observation laws for two actual curves with ordinary, absolutely simple Jacobians. It combines a binomial comparison with a uniform bound on the closed-point tail. Thus access to the full observations does not improve the sample order attained by the phase estimator. The result concerns the information in a specified observation channel. It does not assume an efficient implementation of that channel.
Haonan Gao, Daeyoung Ham, Yifei Zhang, Xinyuan Tian, Shengxian Ding, Yize Zhao, et al.
Fri 13 Nov
stat.ME
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.69
- interest Kernels, operators and functional data 0.71
- favorites near “Why Does Deep Learning Work? - A Perspective From Group Theory” 0.73
Abstract
Characterizing the relationship between brain anatomical wiring and functional coordination remains a fundamental challenge in computational neuroscience. This difficulty primarily arises because structure-function coupling in the human brain is spatially heterogeneous, often localized to subnetworks, and inherently difficult to align across disparate representational modalities. We introduce Orthogonal Sparse Subgraph Alignment (OSSA), a principled and interpretable framework designed to identify compact structural cliques that exhibit maximal concordance with functional connectivity (FC) across subjects. Within each FC-derived community, OSSA optimizes an alignment objective over an orthogonal transformation, which absorbs coordinate mismatch between the two connectivity modalities, and a sparse node-selection vector, which identifies the induced structural connectivity (SC) subgraph. Furthermore, for canonical OSSA within a fixed population community, we establish local statistical guarantees under the required assumptions. With high probability, an empirical local maximizer exists in a specified neighborhood of the population target, and every local maximizer in that neighborhood exactly recovers the active support and satisfies a nonasymptotic error bound for the aligned functional projector. Extensive empirical evaluations on simulated data, alongside analyses of the Adolescent Brain Cognitive Development (ABCD) baseline and longitudinal cohorts, demonstrate that OSSA successfully recovers aligned structural cliques under latent cross-modal rotations. Ultimately, the proposed methodology significantly improves out-of-sample SC-FC alignment over size-matched baselines and identifies compact structural supports with longitudinal consistency in node selection, while multi-support diagnostics suggest a dominant aligned core within the estimated FC communities.
Muhammad Shoaib Ayub, Rana Muhammad Subhan, Insoo Koo, Felipe Augusto Pereira de Figueiredo
Fri 13 Nov
cs.ITeess.SPmath.ITmath.STstat.TH
pdf html
- reading gains, extremely, state, fewer
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.66
- favorites near “Monte Carlo Geometry Processing” 0.70
Abstract
Spatial non-stationarity in multi-cell extremely large-scale MIMO (XL-MIMO) confines each user to a visibility region (VR) of each array. VR-Net exploits these regions for subarray-level coordinated beamforming. A base-station (BS)-indexed conflict graph identifies overlapping VRs; gain-based association and a masked multi-cell weighted minimum mean square error (WMMSE) iteration restrict beam design and channel state information (CSI) uploads to the relevant subarray blocks. We establish conditions for exact per-cell decoupling, derive the expected CSI loads, bound the loss of designing on the VR-gated channel and the loss of the support constraint, and characterize single-edge visibility contrast through a closed-form Fresnel transition width. Simulations use a 3GPP TR 38.901 Release-19-based near-field UMi channel with modular facade arrays and pre-registered criteria. At a 30 dB threshold, VR-Net retains at least 99.6% of the full-aperture WMMSE sum rate in every tested setting while uploading 10-39% fewer complex CSI entries. It gains up to 3.5 bit/s/Hz over an all-user adaptation of the multi-cell scheme of P\'erez-Ad\'an et al. at moderate power and dense load, and inter-cell coordination adds up to 33% over per-cell design.
Mario Edoardo Pandolfo, Kyriakos Stylianopoulos, George C. Alexandropoulos, Paolo Di Lorenzo
Tue 17 Nov
eess.SPcs.LGstat.ML
pdf html
- favorites near “Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss” 0.74
- interest Language models as statistical objects 0.66
- your work near “Computational Methods for Economic Models with Function Valued States” 0.59
Abstract
Semantic communications enable task-oriented information exchange, but suffer from latent-space mismatch when transmitter and receiver employ heterogeneous machine learning models. Existing semantic alignment methods rely on digital processing, introducing computational and latency overhead. This paper proposes a dual stacked intelligent metasurfaces (dual-SIM) architecture that performs semantic alignment entirely in the wave domain. By jointly exploiting programmable metasurfaces at both ends of a link and the wireless channel, the proposed architecture realizes an end-to-end linear transformation between heterogeneous latent spaces. We show phase-only multilayer SIM can approximate both supervised linear aligners and zero-shot Parseval-frame equalizers. We investigate single-SIM, disjoint asymmetric and symmetric dual-SIM, and jointly optimized dual-SIM configurations, considering channel equalization and constructive channel exploitation. For joint design, we develop an alternating projected gradient algorithm with convergence guarantees to first-order stationary points. We further introduce a deep-unrolled network that unrolls gradient-based iterations to reduce computational cost of SIM reconfiguration under channel variations. Numerical results with heterogeneous vision transformer encoders show accurate semantic alignment from moderate to high signal-to-noise ratios. The joint and symmetric disjoint dual-SIM designs achieve the highest accuracy over compact channels, the joint design leads at shallow stack depths, and the advantage over single-SIM designs diminishes at equal depth.
Khadija Ardif, Meryeme Ayache, Nidal Nasser, Ouiame Marnissi
Fri 13 Nov
stat.MLcs.LG
pdf html
- your work near “Online Data Collection for Efficient Semiparametric Inference” 0.72
- interest Language models as statistical objects 0.64
- favorites near “Hypoelliptic diffusion maps I: tangent bundles” 0.73
Abstract
Federated multi-view clustering (FedMVC) has recently attracted increasing attention for its ability to exploit complementary clustering structures across multiple clients while preserving data privacy. However, existing FedMVC methods often suffer from high communication overhead due to the repeated transmission of high-dimensional representations, which limits their scalability in practical federated systems. Moreover, different views may contribute unequally to the clustering performance, calling for adaptive communication strategies. In this paper, we propose a view-aware quantization algorithm for communication-efficient FedMVC. By dynamically assigning quantization bits based on iteratively updated view importance scores, the proposed algorithm effectively balances clustering performance and communication efficiency under a global bit budget constraint. To validate its generality, we integrate our approach into four state-of-the-art FedMVC methods. Extensive experiments on multiple benchmark datasets demonstrate that view-aware quantization achieves up to 6.59x communication reduction while maintaining and in some cases surpassing baseline clustering performance, confirming the effectiveness and scalability of the proposed approach.
Awni Altabaa, John Lafferty, Omar Montasser
Tue 17 Nov
cs.LGstat.ML
pdf html
- reading vc, present algorithm, distilled, multiclass
- favorites near “Sequential Monte Carlo Steering of Large Language Models using Probabilistic Programs” 0.76
- interest Language models as statistical objects 0.75
- looks like incremental deep learning
Abstract
We study how to learn an autoregressive next-token generator $f\colon\Sigma^\star\to\Sigma$ from a class $\mathcal{F}$, given chain-of-thought (CoT) demonstrations of length $T$ generated by $f$. We present a simple algorithm based on bagging, which outputs a majority vote of generators that are trained to be CoT-consistent on bootstrap samples of the training data. For binary token sets, it achieves the optimal sample complexity of $O((d_{\mathrm{VC}}+\log(1/\delta))/\epsilon)$, independent of the CoT length $T$, where $d_{\mathrm{VC}}$ is the VC dimension of $\mathcal{F}$. For finite token sets $\Sigma$, the VC dimension is replaced by a multiclass analogue, the Graph dimension $d_{\mathrm{G}}$, and the sample complexity is optimal up to a $\log|\Sigma|$ factor. The number of generators in the majority-vote ensemble is $O(\log(T/\delta))$, and we show that the ensemble can be distilled into a single generator in $\mathcal{F}$ at the expense of $n\log(nT)$ unlabeled prompts.
Wei Wei, Liangze Ke, Tianzhi Tang, Pavan Challa, Tulika Rudra Gupta, Guannan Gong, et al.
Fri 13 Nov
stat.ME
pdf html
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.77
- favorites near “Bayesian Semiparametric Model for Sequential Treatment Decisions with Informative Timing” 0.82
- interest Empirical Bayes, shrinkage and decision theory 0.79
- looks like clinical and biomedical studies
Abstract
External trial evidence can inform oncology trial design, but broad pooling may produce benchmarks poorly aligned with the planned population. We developed TRACE (Target-Trial Relevance Assessment and Compatibility-Guided Evidence Synthesis), combining eligibility screening, Bayesian compatibility assessment using published baseline characteristics, and hierarchical survival modeling. An integrated compatibility score based on the largest standardized difference in jointly reported characteristics identifies compatible external trials. Their reconstructed control-arm survival data are synthesized using a hierarchical beta-Stacy process to generate a predictive survival benchmark. We evaluated TRACE using eleven randomized phase III trials in first-line HER2-negative advanced gastric/gastroesophageal junction adenocarcinoma. With KEYNOTE-859 as the target, three external trials were selected. TRACE predicted a control-arm median overall survival of 11.8 months (95% posterior predictive interval, 8.9-15.2), compared with an observed median of 11.8 months (95% confidence interval, 10.9-12.7). In leave-one-trial-out validation, seven targets had compatible external evidence. Predictive intervals contained the observed median for all seven. In comparison, confidence intervals from Kaplan-Meier estimates pooling all other trials' control arms contained it for five. Mean pointwise curve coverage was 92% for TRACE and 59% for pooled Kaplan-Meier. For four targets without compatible external evidence, TRACE returned a weakly informative prior predictive benchmark. TRACE provides a transparent approach to selecting relevant external evidence and generating benchmarks for oncology trial planning.
Francesco Cellitti, Lorenzo Di Rocco, Livia Perfetto, Veronica Lombardo, Umberto Ferraro Petrillo
Fri 13 Nov
stat.APcs.LGq-bio.MNstat.ML
pdf html
- interest Kernels, operators and functional data 0.73
- your work near “Local Causal Discovery for Estimating Causal Effects” 0.70
- reading generic, functional, kernels, research
- looks like clinical and biomedical studies
Abstract
The integration of mechanistic signaling networks into clinical research has become a cornerstone of precision oncology. By capturing the complex biochemical interactions within a cell, these networks provide a functional map of disease progression that goes beyond simple genomic mutations. Consequently, the development of specific signaling models of patients has significant relevance, allowing for a more granular representation of individual disease profiles. In this setting, there is a growing need for robust methodologies capable of comparing graph based pathway structures to identify similarities and differences across patient cohorts. Such comparisons are essential for the identification of clinically meaningful subgroups, and the discovery of novel biomarkers. Standard graph kernels, while mathematically rigorous, often remain too generic for this purpose. They frequently fail to capture features specific to gene signaling networks, resulting in suboptimal clustering performance. In this paper, we present a domain specific graph kernel designed for mechanistic signaling networks. Unlike generic approaches, our method explicitly incorporates biological constraints and gene interaction patterns into the kernel formulation, enabling a more accurate characterization of both topological structure and functional behavior. We assess its performance on breast cancer datasets, focusing on its ability to support disease subtyping. Our results suggest that our approach can effectively contribute to patient stratification.