Statistical Machine Learning
Statistical approaches to machine learning, Bayesian methods, and theoretical foundations.
stat.ML · 377 papersTransformers Efficiently Perform In-Context Logistic Regression via Normalized Gradient Descent
Transformers can efficiently perform in-context logistic regression through layers that mimic normalized gradient descent steps.
DARTS: Targeting Prognostic Covariates in Budget-Constrained Sequential Experiments
DARTS is a new method for budget-constrained sequential experiments that adaptively selects prognostic covariates to improve treatment effect estimation efficiency.
A Geometry-Aware Residual Correction of Hagan's SABR Implied Volatility Formula
This paper proposes a geometry-aware neural network to correct Hagan's SABR implied volatility formula, improving accuracy and robustness.
Dynamic Treatment on Networks
Q-Ising is a new pipeline for dynamic treatment allocation on networks, combining Bayesian Ising models with offline reinforcement learning.
Hedging Memory Horizons for Non-Stationary Prediction via Online Aggregation
MELO is an online aggregation method that hedges memory horizons to adapt to non-stationary data, outperforming baselines in electricity load forecasting.
Estimate Level Adjustment For Inference With Proxies Under Random Distribution Shifts
A new estimate-level framework calibrates proxy-based inference under random distribution shifts, avoiding strict assumptions and individual-level data.
Risk-Controlled Post-Processing of Decision Policies
This paper introduces risk-controlled post-processing for decision policies, maximizing agreement with baselines under specified risk constraints.
Q-MMR: Off-Policy Evaluation via Recursive Reweighting and Moment Matching
Q-MMR is a novel off-policy evaluation framework using recursive reweighting and moment matching, offering dimension-free finite-sample guarantees.
Neural-Actuarial Longevity Forecasting: Anchoring LSTMs for Explainable Risk Management
This paper introduces Hybrid-Lift, a neural-actuarial framework combining LSTMs with MBC to improve longevity forecasting, especially in high-longevity countries.
Decoupled PFNs: Identifiable Epistemic-Aleatoric Decomposition via Structured Synthetic Priors
Decoupled PFNs use structured synthetic priors to identifiably separate epistemic and aleatoric uncertainty, improving sequential decision-making in noisy settings.
Covariate Balancing and Riesz Regression Should Be Guided by the Neyman Orthogonal Score in Debiased Machine Learning
This paper argues that debiased machine learning should use regressor balancing guided by the Neyman orthogonal score, not just covariate balancing.
Beyond the Independence Assumption: Finite-Sample Guarantees for Deep Q-Learning under $τ$-Mixing
This paper provides finite-sample guarantees for Deep Q-learning by explicitly modeling temporally dependent replay data as $\tau$-mixing.
The Interplay of Data Structure and Imbalance in the Learning Dynamics of Diffusion Models
This paper analyzes how data structure and imbalance affect diffusion model learning, showing class variance and sampling bias dictate generalization and memorization.
Order-Agnostic Autoregressive Modelling with Missing Data
This paper introduces MO-ARM, an order-agnostic autoregressive model for robust imputation and active information acquisition in datasets with missing data.
Topological Signatures of Grokking
Persistent homology identifies a clear topological signature of grokking, showing a sharp increase in H1 persistence linked to generalization.
TinyBayes: Closed-Form Bayesian Inference via Jacobi Prior for Real-Time Image Classification on Edge Devices
TinyBayes offers a novel closed-form Bayesian classifier for real-time crop disease detection on edge devices, combining small models with uncertainty quantification.
End-to-End Identifiable and Consistent Recurrent Switching Dynamical Systems
This paper introduces ΩSDS, a flow-based estimator for identifiable recurrent switching dynamical systems, improving disentanglement and forecasting.
Attributions All the Way Down? The Metagame of Interpretability
The 'metagame' framework quantifies second-order interaction effects in model explanations using Shapley values, providing deeper interpretability.
Optimal Contextual Pricing under Agnostic Non-Lipschitz Demand
This paper introduces a new algorithm for contextual dynamic pricing with non-Lipschitz demand, achieving optimal \tilde O(T^{2/3}) regret.
Sharp Capacity Thresholds in Linear Associative Memory: From Winner-Take-All to Listwise Retrieval
This paper reveals sharp capacity thresholds for linear associative memories, showing $n \log n$ for winner-take-all and $n$ for listwise retrieval.
📬 Weekly AI Paper Digest
Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.