Methodology

How it works

Every weekday night, after arXiv's 20:00 US-Eastern announcement, a GitHub Actions job fetches the papers announced in the last 7 days of listings in stat.ML, stat.ME, math.ST, stat.AP, stat.CO, stat.OT, econ.EM (new submissions, including cross-lists, and papers released late from moderation, dated by the listing they actually appeared in), scores each one on four criteria against a profile, and rebuilds this site. Nothing is filtered out: irrelevant papers are just ranked low.

The four criteria

Each paper's title and abstract is embedded with sentence-transformers/allenai-specter, a small model that runs on CPU. Each criterion is then converted to the paper's rank within the run, scaled to [0, 1], because raw similarities against different sets live on different scales. The ranks are combined with these weights:

criterion	what it asks	weight
your work	weighted maximum cosine similarity to one of my papers	0.3
favorites	weighted maximum cosine similarity to a paper I've listed as a favorite	0.2
interests	weighted maximum cosine similarity to one of the interest statements below	0.25
reading	how much the paper looks like my whole Zotero library (2489 papers read since grad school): mostly a tf-idf logistic-regression classifier trained to tell my library from a background sample of 15380 arXiv papers in these categories, plus a smaller part for mean similarity to the 10 nearest library papers (held out, the classifier tells arXiv preprints I've read from ones I haven't with AUC 0.8909)	0.25

A paper that is closer to one of the avoid statements than to any of my papers, favorites or interests loses up to 0.3 from its combined score. It is pushed down, not removed.

Before the best match is taken, each profile item's average similarity to that day's papers is subtracted, so a generic item that is moderately close to everything doesn't win every comparison.

The highlighted sections at the top show only papers from the latest listing. Each paper appears under the one criterion it ranks highest on, so the sections don't repeat each other. The "why" line under every paper names the profile item that matched best, with its cosine similarity. For the reading criterion it shows the terms that pushed the classifier's score up the most.

Only derived data from the library is published: the classifier's vocabulary and coefficients, and embedding vectors. Titles and abstracts of the library stay on my laptop.

Interests

Semiparametric inference and debiased machine learning weight 1.5

Semiparametric efficiency and estimation of low-dimensional functionals with machine-learned nuisance parameters: influence functions and efficient influence function calculus, double or debiased machine learning, orthogonal and Neyman-orthogonal moments, cross-fitting, doubly robust and targeted estimators, automatic debiasing through Riesz representers and Riesz regression, automatic differentiation of influence functions, kernel debiased plug-in estimation, and inference for conditional or heterogeneous effects built on the same machinery. Includes prediction-powered and design-based semi-supervised inference, where model predictions, LLM annotations or synthetic data stand in for missing labels and validity is restored with a small gold-standard sample.

Adaptive data collection and anytime-valid inference weight 1.5

Statistical inference when data are collected adaptively or sequentially: adaptive experimental design, bandit algorithms used for estimation rather than regret, active and online data collection to estimate a target parameter efficiently, deciding which data source or measurement to acquire under a budget, inference after adaptive experiments, martingale concentration, confidence sequences, e-values and e-processes, safe anytime-valid and game-theoretic inference, and design-based confidence sequences for online experiments.

Online learning and statistical learning theory weight 1.2

Theory of prediction and learning: online learning and regret minimization against adversarial or nonstationary data, sequential Rademacher complexity, sequential probability assignment and minimax regret under log loss, relaxations and algorithm design from minimax analysis, eluder dimension and decision-estimation coefficients for interactive decision making, and generalization theory for overparameterized models — double descent, benign overfitting and interpolation, PAC-Bayes and compression bounds, algorithmic stability.

Reinforcement learning, policy learning and off-policy evaluation weight 1.2

Sequential decision making viewed statistically: off-policy evaluation and its semiparametric efficiency, doubly robust and double reinforcement learning, dynamic treatment regimes and optimal treatment allocation, policy learning with welfare or regret guarantees, offline reinforcement learning, sample complexity of model-based RL, costly observation and timing of actions, and reinforcement learning methods for solving economic models and finding equilibria.

Causal inference and identification in econometrics weight 1.2

Econometric methods for causal effects: identification with instruments, marginal treatment effects and continuous instruments, mediation, front-door and proximal identification, causal discovery and graphical models used to choose adjustment sets, difference-in-differences and event studies with staggered or continuous treatment, synthetic control, partial identification and inference on identified sets, weak identification, and causal interpretation of time series estimands such as impulse responses, local projections and structural VARs.

Computational methods for heterogeneous agent and dynamic economic models weight 1.3

Numerical solution of dynamic stochastic general equilibrium models with heterogeneous agents and aggregate shocks: HANK, Aiyagari and Krusell-Smith economies, sequence-space Jacobian methods, perturbation and linearization in function space, projection methods and sparse grids, continuous-time models solved as systems of partial differential equations, mean field games, and deep learning and neural network solution methods for high-dimensional dynamic programs, including physics-informed networks, deep BSDE solvers and neural operators.

Bayesian computation and structural estimation weight 1.2

Estimation of structural economic and statistical models by simulation and sampling: Hamiltonian Monte Carlo, Langevin dynamics and their convergence theory, sampling as optimization through gradient flows and optimal transport, sequential Monte Carlo and particle filters, variational inference and its behaviour under misspecification, differentiable state-space models and probabilistic programming, simulation-based and likelihood-free inference, moment matching and indirect inference, Bayesian workflow and model checking, and Bernstein-von Mises theorems.

Macroeconometrics and time series weight 1.0

Econometrics for macroeconomic time series: forecasting and nowcasting with large or shifting data, Bayesian VARs and priors for long-run relationships, low-frequency econometrics, local projections versus VARs, identification of structural shocks and impulse responses, distributional and heterogeneous responses to aggregate shocks, and combining micro and macro data to identify aggregate effects of monetary and fiscal policy.

Kernels, operators and functional data weight 1.0

Learning with infinite-dimensional objects: reproducing kernel Hilbert spaces and kernel mean embeddings, Koopman and transfer operators for dynamical systems, operator learning and neural operators, functional data regression and functional instrumental variables, ill-posed inverse problems and nonparametric IV, and spectral methods — graph Laplacians, diffusion maps and manifold learning.

Language models as statistical objects weight 1.0

Statistical and theoretical views of large language models and deep learning: in-context learning as statistical estimation, transformers as algorithms, probabilistic inference and sequential Monte Carlo over language model outputs, preference learning, reward modelling and dueling bandits behind RLHF, using LLM outputs and simulations as data for social science with valid inference, and results that change how the field understands why deep learning generalizes.

Empirical Bayes, shrinkage and decision theory weight 1.0

Compound decision problems and empirical Bayes: nonparametric maximum likelihood for mixing distributions, shrinkage and hierarchical models for many parallel estimates, meta-analysis and aggregating evidence across studies, forecasting with dynamic panels, and statistical decision theory for policy choice under uncertainty.

Things to push down

Incremental deep learning weight 1.0

A new neural network architecture, training trick, prompting method, agent framework or retrieval-augmented generation pipeline evaluated by accuracy on standard benchmarks, with state-of-the-art results and ablations but no statistical, causal or economic question behind it.

Domain applications of machine learning weight 1.0

Applying deep learning or standard machine learning to a specific engineering or science domain: medical imaging, remote sensing, wireless networks, traffic and energy load forecasting, fault detection, materials and molecules, genomics, clinical prediction models, speech and computer vision datasets.

Clinical and biomedical studies weight 1.0

Reports of clinical trials, epidemiological cohort studies, biostatistics for a clinical audience, survival analysis of a particular patient population, pharmacometrics and bioinformatics pipelines.

My papers

Valid Inference with Imperfect Synthetic Data 1.5
Online Data Collection for Efficient Semiparametric Inference 1.5
Efficient Online Estimation of Causal Effects by Deciding What to Observe 1.4
Timing as an Action: Learning When to Observe and Act 1.2
Local Causal Discovery for Estimating Causal Effects 1.2
Estimating Treatment Effects with Observed Confounders and Mediators 1.2
Differentiable State Space Models and Hamiltonian Monte Carlo Estimation 1.3
Estimating Distributional Responses to Macroeconomic Shocks 1.0
Automated Solution of Heterogeneous Agent Models 1.2
Solution of Rational Expectations Models with Function Valued States 1.0
Perturbation Methods for Incomplete Markets Economies 1.0
Computational Methods for Economic Models with Function Valued States 0.8

Favorites

From my year-end "papers I liked" posts and papers I've linked on Bluesky.

Cause, Effect, and the Structure of the Social World 1.0
Spectral State Space Models 1.0
A U-turn on Double Descent: Rethinking Parameter Counting in Statistical Learning 1.0
Classical Statistical (In-Sample) Intuitions Don't Generalize Well: A Note on Bias-Variance Tradeoffs, Overfitting and Moving from Fixed to Random Designs 0.7
Price Level and Inflation Dynamics in Heterogeneous Agent Economies 1.0
Fiscal Histories 0.7
Empirical Bayes for the Reluctant Frequentist 1.0
Simplifying debiased inference via automatic differentiation and probabilistic programming 1.3
Semiparametric doubly robust targeted double machine learning: a review 0.7
The Minimax Regret of Sequential Probability Assignment, Contextual Shtarkov Sums, and Contextual Normalized Maximum Likelihood 1.0
Revisiting the Phillips and Beveridge Curves: Insights from the 2020s Inflation Surge 1.0
Finding Regularized Competitive Equilibria of Heterogeneous Agent Macroeconomic Models via Reinforcement Learning 1.0
The Trouble with Rational Expectations in Heterogeneous Agent Models: A Challenge for Macroeconomics 0.7
Time-uniform central limit theory and asymptotic confidence sequences 0.7
Eluder Dimension and the Sample Complexity of Optimistic Exploration 1.0
Foundations of Reinforcement Learning and Interactive Decision Making 0.7
From Optimization to Sampling Through Gradient Flows 1.0
When Did Growth Begin? New Estimates of Productivity Growth in England from 1250 to 1870 1.0
Neural Operator: Learning Maps Between Function Spaces With Applications to PDEs 1.0
Fit without fear: remarkable mathematical phenomena of deep learning through the prism of interpolation 1.0
PAC-Bayes Compression Bounds So Tight That They Can Explain Generalization 0.7
The No Free Lunch Theorem, Kolmogorov Complexity, and the Role of Inductive Biases in Machine Learning 0.7
Recent Developments in Machine Learning Methods for Stochastic Control and Games 1.0
Using Imperfect Surrogates for Downstream Inference: Design-based Supervised Learning for Social Science Applications of Large Language Models 1.0
Cross-Prediction-Powered Inference 0.7
More power to you: Using machine learning to augment human coding for more efficient inference in text-based randomized trials 0.7
Sequential Monte Carlo Steering of Large Language Models using Probabilistic Programs 1.0
The Consensus Game: Language Model Generation via Equilibrium Search 0.7
Data Science at the Singularity 1.0
Preference-based Online Learning with Dueling Bandits: A Survey 1.0
Majorizing Measures, Sequential Complexities, and Online Learning 1.0
Sequential complexities and uniform martingale laws of large numbers 0.7
Eigendecompositions of Transfer Operators in Reproducing Kernel Hilbert Spaces 1.0
Applied Koopmanism 0.7
Advances in Nowcasting Economic Activity: Secular Trends, Large Shocks and New Data 1.0
Measuring Price Selection in Microdata: It's Not There 1.0
The Missing Intercept: A Demand Equivalence Approach 0.7
Identification in Macroeconomics 0.7
Hicks-Arrow Prices for US Federal Debt 1791-1930 1.0
Difference-in-Differences with multiple time periods 1.0
Deep Learning for Individual Heterogeneity 1.0
Demystifying statistical learning based on efficient influence functions 0.7
Orthogonal Statistical Learning 1.0
Double/debiased machine learning for treatment and structural parameters 0.7
When do common time series estimands have nonparametric causal meaning? 1.0
Labor Rationing 1.0
Dynamically Optimal Treatment Allocation 1.0
Optimal Decision Rules for Weak GMM 1.0
Distributional Random Forests: Heterogeneity Adjustment and Multivariate Distributional Regression 1.0
Bayesian Workflow 1.0
Priors for the Long Run 1.0
Double Reinforcement Learning for Efficient Off-Policy Evaluation in Markov Decision Processes 1.0
Statistically Efficient Off-Policy Policy Gradients 0.7
Doubly Robust Off-Policy Value and Gradient Estimation for Deterministic Policies 0.7
Monte Carlo Geometry Processing 1.0
Eight Centuries of Global Real Interest Rates, R-G, and the 'Suprasecular' Decline, 1311-2018 1.0
SVAR (Mis)Identification and the Real Effects of Monetary Policy Shocks 1.0
Variational Bayes under Model Misspecification 1.0
Earnings and Consumption Dynamics: A Nonlinear Panel Data Framework 0.8
Policy Learning with Observational Data 0.8
Faster Rates for Policy Learning 0.6
A Machine Learning Approach to Optimal Policy and Taxation 0.8
Honest confidence sets in nonparametric IV regression and other ill-posed models 0.8
Geometric MCMC for Infinite-Dimensional Inverse Problems 0.8
A Conceptual Introduction to Hamiltonian Monte Carlo 0.6
MCMC Confidence Sets for Identified Sets 0.8
Aggregating Distributional Treatment Effects: A Bayesian Hierarchical Analysis of the Microcredit Literature 0.8
Long-Run Covariability 0.8
Bayesian Inference on Structural Impulse Response Functions 0.8
Learning Scalable Deep Kernels with Recurrent Structure 0.8
Variational Inference: A Review for Statisticians 0.6
Hierarchies of Relaxations for Online Prediction Problems with Evolving Constraints 1.0
Why Does Deep Learning Work? - A Perspective From Group Theory 0.8
Random Walks on Simplicial Complexes and Harmonics 0.8
Operator-valued Kernels for Learning from Functional Response Data 0.8
Introduction to regularity structures 0.8
Estimation with Aggregate Shocks 0.8
Mean field games and applications 0.8
Hypoelliptic diffusion maps I: tangent bundles 0.8
Instrumental variable estimation in functional linear models 0.8
A survey of sequential Monte Carlo methods for economics and finance 0.8
Optimal uniform convergence rates and asymptotic normality for series estimators under weak dependence and weak conditions 0.8
Uniform post-selection inference for least absolute deviation regression and other Z-estimation problems 0.8
The Geography of Development: Evaluating Migration Restrictions and Coastal Flooding 0.8
Macroeconomics and Household Heterogeneity 0.7
The Dynamics of Inequality 0.7
Monetary Policy According to HANK 0.7
Monetary Policy and the Redistribution Channel 0.7
From Continuous Dynamics to Practical Gradient-Based Samplers 0.6
A Coreset Selection of Coreset Selection Literature 0.6
The Incredible Flexibility of Moment Matching 0.6
Indirect inference and calibration of dynamic stochastic general equilibrium models 0.6
RieszBoost: Gradient Boosting for Riesz Regression 0.6
Automatic Debiased Machine Learning of Causal and Structural Effects 0.6
Design-Based Confidence Sequences: A General Approach to Risk Mitigation in Online Experimentation 0.6
Positive Long-Run Capital Taxation: Chamley-Judd Revisited 0.6
The Intertemporal Keynesian Cross 0.6
Game-Theoretic Statistics and Safe Anytime-Valid Inference 0.6
Debiased Machine Learning U-statistics 0.6
Cross-Fitting-Free Debiased Machine Learning with Multiway Dependence 0.6
AI4SLT: Empirical Processes in Lean 4 for Formal Statistical Learning Theory 0.6
Visually Communicating and Teaching Intuition for Influence Functions 0.6
Economic Predictions with Big Data: The Illusion of Sparsity 0.6
Underspecification Presents Challenges for Credibility in Modern Machine Learning 0.6
Reinforcement Learning: An Overview 0.6
Higher Order Kernel Mean Embeddings to Capture Filtrations of Stochastic Processes 0.6
Honest Inference for Stochastic Optimization 0.6
Bernstein-von Mises theorems for time evolution equations 0.6
Physics-informed machine learning: A mathematical framework with applications to time series forecasting 0.6
Strategizing against No-regret Learners 0.6
Forecasting with Dynamic Panel Data Models 0.6
Online Learning via Sequential Complexities 0.6
Minimax rates for heterogeneous causal effect estimation 0.6
Can Machines Learn Weak Signals? 0.6
An Invitation to Sequential Monte Carlo Samplers 0.6
Inference with Hamiltonian Sequential Monte Carlo Simulators 0.6
Anytime-Valid Inference for Double/Debiased Machine Learning of Causal Parameters 0.6
Streamlined mean field variational Bayes for longitudinal and multilevel data analysis 0.6
Learning Rich Rankings 0.6
Costs of Financing U.S. Federal Debt Under a Gold Standard: 1791-1933 0.6
Why Do People Stay Poor? 0.6
Efficiency of Weighted Average Derivative Estimators and Index Models 0.6
Robust causal inference with continuous instruments using the local instrumental variable curve 0.6
A Practical Introduction to Bayesian Estimation of Causal Effects: Parametric and Nonparametric Approaches 0.6
Bayesian Semiparametric Model for Sequential Treatment Decisions with Informative Timing 0.6
From LATE to ATE: A Bayesian approach 0.6
Train faster, generalize better: Stability of stochastic gradient descent 0.6
Survival Analysis via Ordinary Differential Equations 0.6
Stabilized Neural Prediction of Potential Outcomes in Continuous Time 0.6
Graphical criteria for the identification of marginal causal effects in continuous-time survival and event-history analyses 0.6
Post-selection inference for causal effects after causal discovery 0.6
The Best of Both Worlds: Combining Randomized Controlled Trials with Structural Modeling 0.6
Causal mediation analysis with double machine learning 0.6
Posterior distribution of nondifferentiable functions 0.6
Overidentification in Regular Models 0.6
Central Limit Theorems for Smooth Optimal Transport Maps 0.6
Multivariate Rank-Based Distribution-Free Nonparametric Testing Using Measure Transportation 0.6
Minimax Confidence Intervals for the Sliced Wasserstein Distance 0.6
Deep Learning for Economists 0.6
Adapting Text Embeddings for Causal Inference 0.6
Potential Outcome and Directed Acyclic Graph Approaches to Causality: Relevance for Empirical Practice in Economics 0.6
Kernel Debiased Plug-in Estimation: Simultaneous, Automated Debiasing without Influence Functions for Many Target Parameters 0.6
The influence function of semiparametric estimators 0.6
Transformers as Statisticians: Provable In-Context Learning with In-Context Algorithm Selection 0.6
Universal inference 0.6
Towards a Complete Analysis of Langevin Monte Carlo: Beyond Poincaré Inequality 0.6
Heavy-tailed Sampling via Transformed Unadjusted Langevin Algorithm 0.6
Estimating Vector Fields on Manifolds and the Embedding of Directed Graphs 0.6
CAREER: A Foundation Model for Labor Sequence Data 0.6
Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss 0.6
Who Should Be Treated? Empirical Welfare Maximization Methods for Treatment Choice 0.6
Generalization of GMM to a Continuum of Moment Conditions 0.6
Consistent Estimation of Models Defined by Conditional Moment Restrictions 0.6
Monopsony in Online Labor Markets 0.6
The Wealthy Hand-to-Mouth 0.6

Caveats

This is all machine similarity, and it is sometimes wrong. The ranks are relative to one run, so they compare a paper with the others found that week and say nothing more. Papers first announced in another category and cross-listed into these categories later don't appear.