@inproceedings{wang-etal-2024-reward,
title = "Reward Difference Optimization For Sample Reweighting In Offline {RLHF}",
author = "Wang, Shiqi and
Zhang, Zhengze and
Zhao, Rui and
Tan, Fei and
Cam-Tu, Nguyen",
editor = "Al-Onaizan, Yaser and
Bansal, Mohit and
Chen, Yun-Nung",
booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2024",
month = nov,
year = "2024",
address = "Miami, Florida, USA",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2024.findings-emnlp.115",
pages = "2109--2123",
abstract = "With the wide deployment of Large Language Models (LLMs), aligning LLMs with human values becomes increasingly important. Although Reinforcement Learning with Human Feedback (RLHF) proves effective, it is complicated and highly resource-intensive. As such, offline RLHF has been introduced as an alternative solution, which directly optimizes LLMs with ranking losses on a fixed preference dataset. Current offline RLHF only captures the ordering relationship between responses, overlooking the crucial aspect of {``}how much{''} one is preferred over the others. To address this issue, we propose a simple yet effective solution based on reward difference prediction. Specifically, we introduce reward difference coefficients to reweigh sample pairs in offline RLHF. We then propose a difference model that considers rich interactions between a pair of responses for predicting these difference coefficients. Experiments with 7B LLMs on the HH and TL;DR dataset verify the effectiveness of our method in both automatic metrics and human evaluation, highlighting its potential for aligning LLMs with human values.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2024-reward">
<titleInfo>
<title>Reward Difference Optimization For Sample Reweighting In Offline RLHF</title>
</titleInfo>
<name type="personal">
<namePart type="given">Shiqi</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhengze</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Rui</namePart>
<namePart type="family">Zhao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fei</namePart>
<namePart type="family">Tan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Nguyen</namePart>
<namePart type="family">Cam-Tu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2024-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2024</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yaser</namePart>
<namePart type="family">Al-Onaizan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohit</namePart>
<namePart type="family">Bansal</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yun-Nung</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Miami, Florida, USA</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>With the wide deployment of Large Language Models (LLMs), aligning LLMs with human values becomes increasingly important. Although Reinforcement Learning with Human Feedback (RLHF) proves effective, it is complicated and highly resource-intensive. As such, offline RLHF has been introduced as an alternative solution, which directly optimizes LLMs with ranking losses on a fixed preference dataset. Current offline RLHF only captures the ordering relationship between responses, overlooking the crucial aspect of “how much” one is preferred over the others. To address this issue, we propose a simple yet effective solution based on reward difference prediction. Specifically, we introduce reward difference coefficients to reweigh sample pairs in offline RLHF. We then propose a difference model that considers rich interactions between a pair of responses for predicting these difference coefficients. Experiments with 7B LLMs on the HH and TL;DR dataset verify the effectiveness of our method in both automatic metrics and human evaluation, highlighting its potential for aligning LLMs with human values.</abstract>
<identifier type="citekey">wang-etal-2024-reward</identifier>
<location>
<url>https://aclanthology.org/2024.findings-emnlp.115</url>
</location>
<part>
<date>2024-11</date>
<extent unit="page">
<start>2109</start>
<end>2123</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Reward Difference Optimization For Sample Reweighting In Offline RLHF
%A Wang, Shiqi
%A Zhang, Zhengze
%A Zhao, Rui
%A Tan, Fei
%A Cam-Tu, Nguyen
%Y Al-Onaizan, Yaser
%Y Bansal, Mohit
%Y Chen, Yun-Nung
%S Findings of the Association for Computational Linguistics: EMNLP 2024
%D 2024
%8 November
%I Association for Computational Linguistics
%C Miami, Florida, USA
%F wang-etal-2024-reward
%X With the wide deployment of Large Language Models (LLMs), aligning LLMs with human values becomes increasingly important. Although Reinforcement Learning with Human Feedback (RLHF) proves effective, it is complicated and highly resource-intensive. As such, offline RLHF has been introduced as an alternative solution, which directly optimizes LLMs with ranking losses on a fixed preference dataset. Current offline RLHF only captures the ordering relationship between responses, overlooking the crucial aspect of “how much” one is preferred over the others. To address this issue, we propose a simple yet effective solution based on reward difference prediction. Specifically, we introduce reward difference coefficients to reweigh sample pairs in offline RLHF. We then propose a difference model that considers rich interactions between a pair of responses for predicting these difference coefficients. Experiments with 7B LLMs on the HH and TL;DR dataset verify the effectiveness of our method in both automatic metrics and human evaluation, highlighting its potential for aligning LLMs with human values.
%U https://aclanthology.org/2024.findings-emnlp.115
%P 2109-2123
Markdown (Informal)
[Reward Difference Optimization For Sample Reweighting In Offline RLHF](https://aclanthology.org/2024.findings-emnlp.115) (Wang et al., Findings 2024)
ACL