Saved in:
Bibliographic Details
Main Authors: Bose, Avinandan, Xiong, Zhihan, Chi, Yuejie, Du, Simon Shaolei, Xiao, Lin, Fazel, Maryam
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2504.14439
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915251475185664
author Bose, Avinandan
Xiong, Zhihan
Chi, Yuejie
Du, Simon Shaolei
Xiao, Lin
Fazel, Maryam
author_facet Bose, Avinandan
Xiong, Zhihan
Chi, Yuejie
Du, Simon Shaolei
Xiao, Lin
Fazel, Maryam
contents Personalizing large language models (LLMs) to accommodate diverse user preferences is essential for enhancing alignment and user satisfaction. Traditional reinforcement learning from human feedback (RLHF) approaches often rely on monolithic value representations, limiting their ability to adapt to individual preferences. We introduce a novel framework that leverages low-rank preference modeling to efficiently learn and generalize user-specific reward functions. By representing reward functions in a low-dimensional subspace and modeling individual preferences as weighted combinations of shared basis functions, our approach avoids rigid user categorization while enabling scalability and few-shot adaptation. We validate our method on multiple preference datasets, demonstrating superior generalization to unseen users and improved accuracy in preference prediction tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2504_14439
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LoRe: Personalizing LLMs via Low-Rank Reward Modeling
Bose, Avinandan
Xiong, Zhihan
Chi, Yuejie
Du, Simon Shaolei
Xiao, Lin
Fazel, Maryam
Machine Learning
Artificial Intelligence
Computation and Language
Personalizing large language models (LLMs) to accommodate diverse user preferences is essential for enhancing alignment and user satisfaction. Traditional reinforcement learning from human feedback (RLHF) approaches often rely on monolithic value representations, limiting their ability to adapt to individual preferences. We introduce a novel framework that leverages low-rank preference modeling to efficiently learn and generalize user-specific reward functions. By representing reward functions in a low-dimensional subspace and modeling individual preferences as weighted combinations of shared basis functions, our approach avoids rigid user categorization while enabling scalability and few-shot adaptation. We validate our method on multiple preference datasets, demonstrating superior generalization to unseen users and improved accuracy in preference prediction tasks.
title LoRe: Personalizing LLMs via Low-Rank Reward Modeling
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2504.14439