arXiv AI
Aug 13
Post-Training with Policy Gradients: Optimality and the Base Model Barrier
arXiv:2603. 06957v2 Announce Type: replace-cross Abstract: We study post-training linear autoregressive models with outcome and process rewards.
Tail-Aware Information-Theoretic Bounds for LLM Alignment under Heavy-Tailed Rewards
arXiv:2604. 10727v2 Announce Type: replace-cross Abstract: Classical information-theoretic learning bounds typically rely on KL mutual information and moment-generating-function (MGF) arguments, which are well matched to bounded or sub-Gaussian losses but can be ineffective when losses or rewards are heavy-tailed.
Related stories
arXiv Machine Learning
Sep 24
Type-II Error Bounds for Test Supermartingales from Lower-Tail Hypotheses
arXiv:2609.27766v1 Announce Type: cross Abstract: In safe hypothesis testing with test supermartingals, Ville's inequality provides anytime-valid type-I error guarantees for every significance level...
arXiv Machine Learning
Aug 3
Parameter-Free Heavy-Tailed Bandits
arXiv:2607. 29460v1 Announce Type: new Abstract: Heavy-tailed distributions arise naturally in sequential decision-making problems such as financial investment, online advertising, and network management, where rare but extreme outcomes can dominate performance.
arXiv Machine Learning
4d ago
Fine-Tuning on Self-Generated and Reward-Weighted Data: Learning Dynamics, Convergence Rates, and Benefits of Off-Policyness
arXiv:2609.36945v1 Announce Type: new Abstract: We study the learning dynamics of fine-tuning a policy model on self-generated and reward-weighted data, with particular focus on a generalized version...
arXiv Machine Learning
Jun 9
Asymptotic Optimality of Thompson Sampling for Risk-Averse Bandits with Sub-Gaussian Rewards
arXiv:2606. 09191v1 Announce Type: new Abstract: We prove that $\rho\text{-}\mathrm{NPTS}_{\mathrm{SG}}$, an anchor-free nonparametric Thompson Sampling algorithm for risk-averse bandits, achieves regret matching the instance-dependent lower bound to leading order in $\log n$, establishing it as asymptotically optimal for any continuous risk functional $\rho$ (CVaR, mean-variance, Sharpe ratio, distortion risk measures, and more) on the class of distributions with bounded density and sub-Gaussian tails, including Gaussian arms.
arXiv Machine Learning
Aug 10
Multiscale Reward Hedging from Correct Demonstrations
arXiv:2608. 06825v1 Announce Type: new Abstract: Learning from correct demonstrations is harder than supervised learning when many answers are correct: after predicting, the learner sees one valid answer but not whether its own answer was valid, nor any reward.
arXiv Machine Learning
Jun 9
Noise-Adaptive High-Probability Regret Bounds for Online Convex Optimization
arXiv:2606. 08028v1 Announce Type: new Abstract: We study high-probability regret bounds for online convex optimization (OCO) with strongly convex losses and establish three results that resolve open questions at the intersection of noise adaptivity, feedback structure, and constraint satisfaction.
Hugging Face Trending Papers
Jun 8
Asymptotic Optimality of Thompson Sampling for Risk-Averse Bandits with Sub-Gaussian Rewards
We prove that $ρ\text{-}\mathrm{NPTS}_{\mathrm{SG}}$, an anchor-free nonparametric Thompson Sampling algorithm for risk-averse bandits, achieves regret matching the instance-dependent lower bound to leading order in $\log n$, establishing it as asymptotically optimal for any continuous risk functional $ρ$ (CVaR, mean-variance, Sharpe ratio, distortion risk measures, and more) on the class of distributions with bounded density and sub-Gaussian tails, including Gaussian arms. Both this result and its bounded-support counterpart require only continuity of $ρ$: strictly weaker than the dominance condition of prior parametric Thompson Sampling results, and strictly weaker than the Lipschitz condition of UCB-type algorithms, yielding the first instance-optimal guarantees for non-Lipschitz functionals such as the Sharpe ratio without parametric reward assumptions.
arXiv AI
Aug 25
Safety Hacking in Constrained Best-of-$N$ Inference-time Scaling
arXiv:2608. 22915v1 Announce Type: cross Abstract: Inference-time pipelines often sample multiple outputs, filter them with a learned safety model, and return the proxy-feasible output with the highest learned reward.
arXiv Machine Learning
Sep 14
Distortion of AI Alignment Revisited: RLHF is a Decent Utilitarian Aligner
The paper revisits the distortion problem in Reinforcement Learning from Human Feedback (RLHF) for aligning large language models. It shows that exponential degradation in user utility, previously attributed to RLHF, actually stems from a mismatch between the preference data distribution and the KL reference policy. By deriving tight bounds across different KL regularization regimes, the authors demonstrate that when the two distributions match, RLHF achieves near‑optimal distortion, and they recommend using on‑policy data or pre‑fine‑tuning on data close to the true preference distribution.
arXiv Machine Learning
Aug 11
Kernel Methods for Refined Prophet Inequalities
arXiv:2608. 08662v1 Announce Type: cross Abstract: The single-selection prophet inequality is a canonical Bayesian online selection problem in which independent nonnegative values arrive sequentially and the decision-maker must irrevocably select at most one.
arXiv Statistics ML
3d ago
Tighter Regret Bounds for Contextual Action-Set Reinforcement Learning
arXiv:2605.15692v2 Announce Type: replace-cross Abstract: We study episodic reinforcement learning with fixed reward and transition functions, but with episode-dependent admissible action sets that a...