Statistical Machine Learning
Statistical approaches to machine learning, Bayesian methods, and theoretical foundations.
stat.ML · 377 papersPrice of Quality: Sufficient Conditions for Sparse Recovery using Mixed-Quality Data
This paper provides the first conditions for sparse recovery using mixed-quality data, revealing differences between information-theoretic and algorithmic thresholds.
Natural Policy Gradient as Doubly Smoothed Policy Iteration: A Bellman-Operator Framework
This paper shows Natural Policy Gradient is a doubly smoothed policy iteration, proving its global geometric convergence and optimal complexity.
When Can Digital Personas Reliably Approximate Human Survey Findings?
This paper evaluates when LLM-powered digital personas can reliably substitute human survey respondents, finding they align distributionally but struggle with individual predictions.
Amortizing Causal Sensitivity Analysis via Prior Data-Fitted Networks
An amortized, in-context learning method for causal sensitivity analysis drastically speeds up computation compared to per-instance methods.
Simultaneous Long-tailed Recognition and Multi-modal Fusion for Highly Imbalanced Multi-modal Data
This paper introduces a multi-modal fusion framework for long-tailed recognition in class-imbalanced data, outperforming single-modal methods.
Real vs. Semi-Simulated: Rethinking Evaluation for Treatment Effect Estimation
This paper reveals that common semi-simulated benchmarks and counterfactual metrics for treatment effect estimation don't align with real-world performance.
Sharp feature-learning transitions and Bayes-optimal neural scaling laws in extensive-width networks
This paper reveals sharp feature-learning transitions and Bayes-optimal neural scaling laws in extensive-width networks, unifying two distinct learning regimes.
Regret Analysis of Guided Diffusion for Black-Box Optimization over Structured Inputs
This paper introduces a novel regret analysis framework for guided-diffusion black-box optimization, explaining its strong performance on structured inputs.
Multifidelity Gaussian process regression for solving nonlinear partial differential equations
This paper proposes a multifidelity Gaussian process regression method using cokriging for learning optimal kernels to solve nonlinear PDEs efficiently.
Uncertainty in Physics and AI: Taxonomy, Quantification, and Validation
This paper offers a unified taxonomy, quantification, and validation framework for uncertainty in ML for physics applications.
Quantifying the Risk-Return Tradeoff in Forecasting
This paper introduces a framework to quantify forecast reliability using risk-adjusted financial performance measures, revealing professional forecasters' robust performance.
A Note on Non-Negative $L_1$-Approximating Polynomials
This paper proves the existence of non-negative $L_1$-approximating polynomials for Gaussian distributions, matching optimal degree bounds.
Semiparametric Efficient Test for Interpretable Distributional Treatment Effects
DR-ME is a new semiparametrically efficient test that identifies specific locations where treatment effects alter outcome distributions, unlike global tests.
Penalty-Based First-Order Methods for Bilevel Optimization with Minimax and Constrained Lower-Level Problems
This paper introduces penalty-based first-order methods for bilevel optimization with minimax lower-level problems, achieving improved complexity bounds.
Flow Matching for Count Data
Introduces count-FM, a novel flow-matching framework for high-dimensional count data, achieving better sample quality and efficiency.
Every Feedforward Neural Network Definable in an o-Minimal Structure Has Finite Sample Complexity
Feedforward neural networks definable in o-minimal structures, including MLPs, CNNs, and transformers, possess finite PAC sample complexity.
Causal EpiNets: Precision-corrected Bounds on Individual Treatment Effects using Epistemic Neural Networks
Causal EpiNets use an anchored neural architecture and Epistemic Neural Networks to provide precision-corrected, valid bounds on individual treatment effects.
Response Time Enhances Alignment with Heterogeneous Preferences
This paper shows that using user response times can accurately align LLMs with diverse human preferences, overcoming limitations of standard choice-only methods.
Online Bayesian Calibration under Gradual and Abrupt System Changes
BRPC is an online Bayesian calibration method that adapts to gradual and abrupt system changes, improving accuracy and robustness in digital twin applications.
The Structural Origin of Attention Sink: Variance Discrepancy, Super Neurons, and Dimension Disparity
This paper reveals attention sink origins in LLMs: variance discrepancy, super neurons, and dimension disparity, proposing `head-wise RMSNorm` for mitigation.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.