arrow
返回

Reward-guided direct preference optimization

delete2025-11-04
delete0
PRE
AI
Z
Zhe Yuan Ding
S
Su Pan
Y
Yongpan Zhang
H
Hui Ji
C
Cheng Ding
DOI:10.1016/j.eswa.2025.130295delete
delete原文链接
delete原文求助
delete分享
delete收藏
摘要

摘要

En 中文
• 分析了SFT和DPO中选中与拒绝响应的似然变化。 • 提出了一种动态切换方法以平衡似然与排序目标。 • RG-DPO在偏好学习中优于DPO。 • 提出的算法在HH-RLHF和TLDR数据集上取得了有竞争力的性能。
Keyword:
Alignment
Large language model
Preference optimization
Natural language processing

期刊

Expert Systems with Applications 封面图
Expert Systems with Applications
IF:
7.5
论文数:
3.0W
被引数:
10.2W

机构

N
nanjing university of posts and telecommunications
学者数:
3.8K
论文数: 1.6K
被引数: 0
C
china telecom co, ltd jiangsu branch
学者数:
2
论文数: 1
被引数: 0
引用论文

引用论文

Diffusion Model Alignment Using Direct Preference Optimization
err2024-06-16
err0
PREAI
errBram Wallace; Meihua Dang; Rafael Rafailov; Linqi Zhou; Aaron Lou; Senthil Purushwalkam; Stefano Ermon; Caiming Xiong; Shafiq Joty; Nikhil Naik
err分享
err收藏
TL;DR: Mining Reddit to Learn Automatic Summarization
err2017-01-01
err0
errOAAI
errMichael Völske; Martin Potthast; Shahbaz Syed; Benno Stein
err分享
err收藏