中文
相关论文

相关论文: Towards a Sharp Analysis of Offline Policy Learnin…

200 篇论文

\emph{Kullback-Leibler} (KL) regularization is ubiquitous in reinforcement learning algorithms in the form of \emph{reverse} or \emph{forward} KL. Recent studies have demonstrated $\epsilon^{-1}$-type fast rates for decision making under…

机器学习 · 计算机科学 2026-05-12 Qingyue Zhao , Kaixuan Ji , Heyang Zhao , Quanquan Gu

Reverse-Kullback-Leibler (KL) regularization has emerged to be a predominant technique used to enhance policy optimization in reinforcement learning (RL) and reinforcement learning from human feedback (RLHF), which forces the learned policy…

机器学习 · 计算机科学 2025-02-12 Heyang Zhao , Chenlu Ye , Quanquan Gu , Tong Zhang

Kullback-Leibler (KL) regularization is widely used in offline decision-making and offers several benefits, motivating recent work on the sample complexity of offline learning with respect to KL-regularized performance metrics.…

机器学习 · 计算机科学 2026-05-05 Kaixuan Ji , Qiwei Di , Heyang Zhao , Qingyue Zhao , Quanquan Gu

Recent studies have shown that reinforcement learning with KL-regularized objectives can enjoy faster rates of convergence or logarithmic regret, in contrast to the classical $\sqrt{T}$-type regret in the unregularized setting. However, the…

机器学习 · 计算机科学 2026-03-03 Kaixuan Ji , Qingyue Zhao , Heyang Zhao , Qiwei Di , Quanquan Gu

We study offline reinforcement learning in average-reward MDPs, which presents increased challenges from the perspectives of distribution shift and non-uniform coverage, and has been relatively underexamined from a theoretical perspective.…

机器学习 · 计算机科学 2026-04-23 Matthew Zurek , Guy Zamir , Yudong Chen

Offline reinforcement learning (RL) leverages previously collected data for policy optimization without any further active exploration. Despite the recent interest in this problem, its theoretical results in neural network function…

机器学习 · 统计学 2022-12-15 Thanh Nguyen-Tang , Sunil Gupta , Hung Tran-The , Svetha Venkatesh

We study a generalization of the problem of online learning in adversarial linear contextual bandits by incorporating loss functions that belong to a reproducing kernel Hilbert space, which allows for a more flexible modeling of complex…

机器学习 · 统计学 2023-10-04 Gergely Neu , Julia Olkhovskaya , Sattar Vakili

Offline policy learning (OPL) leverages existing data collected a priori for policy optimization without any active exploration. Despite the prevalence and recent interest in this problem, its theoretical and algorithmic foundations in…

机器学习 · 计算机科学 2022-03-15 Thanh Nguyen-Tang , Sunil Gupta , A. Tuan Nguyen , Svetha Venkatesh

Offline (or batch) reinforcement learning (RL) algorithms seek to learn an optimal policy from a fixed dataset without active data collection. Based on the composition of the offline dataset, two main categories of methods are used:…

机器学习 · 计算机科学 2023-07-04 Paria Rashidinejad , Banghua Zhu , Cong Ma , Jiantao Jiao , Stuart Russell

This work studies the statistical limits of uniform convergence for offline policy evaluation (OPE) problems with model-based methods (for episodic MDP) and provides a unified framework towards optimal learning for several well-motivated…

机器学习 · 计算机科学 2021-06-25 Ming Yin , Yu-Xiang Wang

In this paper, we study the offline and online settings of reinforcement learning from human feedback (RLHF) with KL-regularization -- a widely used objective function in large language model alignment -- under the $\epsilon$ local…

机器学习 · 计算机科学 2025-10-16 Yulian Wu , Rushil Thareja , Praneeth Vepakomma , Francesco Orabona

We study risk-aware offline policy learning, aiming to learn a decision rule from logged data that is optimal under general risk criteria. This problem is crucial in high-stakes domains where online interaction is infeasible and adverse…

机器学习 · 统计学 2026-05-18 Yilong Wan , Yuqiang Li , Xianyi Wu

To ensure stability of learning, state-of-the-art generalized policy iteration algorithms augment the policy improvement step with a trust region constraint bounding the information loss. The size of the trust region is commonly determined…

机器学习 · 计算机科学 2018-04-05 Boris Belousov , Jan Peters

This paper concerns the central issues of model robustness and sample efficiency in offline reinforcement learning (RL), which aims to learn to perform decision making from history data without active exploration. Due to uncertainties and…

机器学习 · 计算机科学 2024-01-01 Laixi Shi , Yuejie Chi

Training generative models to sample from unnormalized density functions is an important and challenging task in machine learning. Traditional training methods often rely on the reverse Kullback-Leibler (KL) divergence due to its…

机器学习 · 计算机科学 2025-03-05 Jiajun He , Wenlin Chen , Mingtian Zhang , David Barber , José Miguel Hernández-Lobato

This paper addresses a new interpretation of the traditional optimization method in reinforcement learning (RL) as optimization problems using reverse Kullback-Leibler (KL) divergence, and derives a new optimization method using forward KL…

机器学习 · 计算机科学 2022-04-25 Taisuke Kobayashi

Sample-efficiency guarantees for offline reinforcement learning (RL) often rely on strong assumptions on both the function classes (e.g., Bellman-completeness) and the data coverage (e.g., all-policy concentrability). Despite the recent…

机器学习 · 计算机科学 2022-06-29 Wenhao Zhan , Baihe Huang , Audrey Huang , Nan Jiang , Jason D. Lee

We address policy learning with logged data in contextual bandits. Current offline-policy learning algorithms are mostly based on inverse propensity score (IPS) weighting requiring the logging policy to have \emph{full support} i.e. a…

机器学习 · 统计学 2021-07-27 Hung Tran-The , Sunil Gupta , Thanh Nguyen-Tang , Santu Rana , Svetha Venkatesh

In offline reinforcement learning, a policy learns to maximize cumulative rewards with a fixed collection of data. Towards conservative strategy, current methods choose to regularize the behavior policy or learn a lower bound of the value…

机器学习 · 计算机科学 2021-03-02 Hongchang Zhang , Jianzhun Shao , Yuhang Jiang , Shuncheng He , Xiangyang Ji

We consider offline policy optimization (OPO) in contextual bandits, where one is given a fixed dataset of logged interactions. While pessimistic regularizers are typically used to mitigate distribution shift, prior implementations thereof…

机器学习 · 计算机科学 2023-10-27 Lequn Wang , Akshay Krishnamurthy , Aleksandrs Slivkins
‹ 上一页 1 2 3 10 下一页 ›