Salvato in:
Dettagli Bibliografici
Autori principali: Lee, Seong Jin, Sun, Will Wei, Liu, Yufeng
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:https://arxiv.org/abs/2412.19436
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908865684045824
author Lee, Seong Jin
Sun, Will Wei
Liu, Yufeng
author_facet Lee, Seong Jin
Sun, Will Wei
Liu, Yufeng
contents Reinforcement learning from human feedback (RLHF) has become a cornerstone for aligning large language models with human preferences. However, the heterogeneity of human feedback, driven by diverse individual contexts and preferences, poses significant challenges for reward learning. To address this, we propose a Low-rank Contextual RLHF (LoCo-RLHF) framework that integrates contextual information to better model heterogeneous feedback while maintaining computational efficiency. Our approach builds on a contextual preference model, leveraging the intrinsic low-rank structure of the interaction between user contexts and query-answer pairs to mitigate the high dimensionality of feature representations. Furthermore, we address the challenge of distributional shifts in feedback through our Pessimism in Reduced Subspace (PRS) policy, inspired by pessimistic offline reinforcement learning techniques. We theoretically demonstrate that our policy achieves a tighter sub-optimality gap compared to existing methods. Extensive experiments validate the effectiveness of LoCo-RLHF, showcasing its superior performance in personalized RLHF settings and its robustness to distribution shifts.
format Preprint
id arxiv_https___arxiv_org_abs_2412_19436
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Low-Rank Contextual Reinforcement Learning from Heterogeneous Human Feedback
Lee, Seong Jin
Sun, Will Wei
Liu, Yufeng
Machine Learning
Reinforcement learning from human feedback (RLHF) has become a cornerstone for aligning large language models with human preferences. However, the heterogeneity of human feedback, driven by diverse individual contexts and preferences, poses significant challenges for reward learning. To address this, we propose a Low-rank Contextual RLHF (LoCo-RLHF) framework that integrates contextual information to better model heterogeneous feedback while maintaining computational efficiency. Our approach builds on a contextual preference model, leveraging the intrinsic low-rank structure of the interaction between user contexts and query-answer pairs to mitigate the high dimensionality of feature representations. Furthermore, we address the challenge of distributional shifts in feedback through our Pessimism in Reduced Subspace (PRS) policy, inspired by pessimistic offline reinforcement learning techniques. We theoretically demonstrate that our policy achieves a tighter sub-optimality gap compared to existing methods. Extensive experiments validate the effectiveness of LoCo-RLHF, showcasing its superior performance in personalized RLHF settings and its robustness to distribution shifts.
title Low-Rank Contextual Reinforcement Learning from Heterogeneous Human Feedback
topic Machine Learning
url https://arxiv.org/abs/2412.19436