$f$-Divergence Regularized RLHF: Two Tales of Sampling and Unified Analyses

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Wu, Di, Shi, Chengshuai, Yang, Jing, Shen, Cong
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866918489073123328
author Wu, Di
Shi, Chengshuai
Yang, Jing
Shen, Cong
author_facet Wu, Di
Shi, Chengshuai
Yang, Jing
Shen, Cong
contents Reinforcement Learning from Human Feedback (RLHF) has become a cornerstone technique for post-training large language models. While most existing approaches rely on the reverse KL-regularization, recent empirical studies have begun exploring alternative divergences (e.g., forward KL, chi-squared) as regularizers in RLHF. However, a unified theoretical understanding of general $f$-divergence regularization remains under-explored. To fill this gap, this work develops a comprehensive theoretical framework for online RLHF with a general $f$-divergence regularized objective. Rather than treating each possible divergence function individually, we adopt a holistic perspective across the entire function class and propose two algorithms based on distinct sampling principles. The first extends the classical optimism principle with a carefully designed exploration bonus, while the second introduces a new method that exploits the sensitivity of the optimal policy to reward perturbations under $f$-divergence regularization. Theoretical analysis shows that $O(\log T)$ regret and $O(1/T)$ sub-optimality gap are achievable, establishing provable efficiency of both algorithms and, to the best of our knowledge, the first performance bounds for online RLHF under general $f$-divergence regularization.
format Preprint
id arxiv_https___arxiv_org_abs_2605_06977
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle $f$-Divergence Regularized RLHF: Two Tales of Sampling and Unified Analyses
Wu, Di
Shi, Chengshuai
Yang, Jing
Shen, Cong
Machine Learning
Artificial Intelligence
Information Theory
Reinforcement Learning from Human Feedback (RLHF) has become a cornerstone technique for post-training large language models. While most existing approaches rely on the reverse KL-regularization, recent empirical studies have begun exploring alternative divergences (e.g., forward KL, chi-squared) as regularizers in RLHF. However, a unified theoretical understanding of general $f$-divergence regularization remains under-explored. To fill this gap, this work develops a comprehensive theoretical framework for online RLHF with a general $f$-divergence regularized objective. Rather than treating each possible divergence function individually, we adopt a holistic perspective across the entire function class and propose two algorithms based on distinct sampling principles. The first extends the classical optimism principle with a carefully designed exploration bonus, while the second introduces a new method that exploits the sensitivity of the optimal policy to reward perturbations under $f$-divergence regularization. Theoretical analysis shows that $O(\log T)$ regret and $O(1/T)$ sub-optimality gap are achievable, establishing provable efficiency of both algorithms and, to the best of our knowledge, the first performance bounds for online RLHF under general $f$-divergence regularization.
title $f$-Divergence Regularized RLHF: Two Tales of Sampling and Unified Analyses
topic Machine Learning
Artificial Intelligence
Information Theory
url https://arxiv.org/abs/2605.06977