SERL: Self-Examining Reinforcement Learning on Open-Domain

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Ou, Weixuan, Zheng, Yanzhao, Sun, Shuoshuo, Zhang, Wei, Dong, Baohua, Zhu, Hangcheng, Huang, Ruohui, Yu, Gang, Yan, Pengwei, Qiao, Yifan
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914349061242880
author Ou, Weixuan
Zheng, Yanzhao
Sun, Shuoshuo
Zhang, Wei
Dong, Baohua
Zhu, Hangcheng
Huang, Ruohui
Yu, Gang
Yan, Pengwei
Qiao, Yifan
author_facet Ou, Weixuan
Zheng, Yanzhao
Sun, Shuoshuo
Zhang, Wei
Dong, Baohua
Zhu, Hangcheng
Huang, Ruohui
Yu, Gang
Yan, Pengwei
Qiao, Yifan
contents Reinforcement Learning (RL) has been shown to improve the capabilities of large language models (LLMs). However, applying RL to open-domain tasks faces two key challenges: (1) the inherent subjectivity of these tasks prevents the verifiable rewards as required by Reinforcement Learning with Verifiable Rewards (RLVR); (2) Reinforcement Learning from Human Feedback (RLHF) relies on external reward mechanisms. To overcome these limitations, we propose Self-Examining Reinforcement Learning (SERL), a novel self-improving framework where the LLM serves as both Actor and Judge. SERL introduces two synergistic reward mechanisms without any external signals. On the one hand, to improve the Actor's capability, we derive rewards from Copeland-style pairwise comparison judgments across a group of generated responses. On the other hand, a self-consistency reward that encourages coherent judgments is proposed to improve the Judge's reliability. This process refines the Judge's capability, which in turn provides a more robust reward for Actor. Experiments show that our method outperforms existing self-improvement training methods. SERL improves the LC win rate of Qwen3-8B on AlpacaEval 2 from 52.37% to 59.90%. To the best of our knowledge, our method achieves state-of-the-art performance among self-improving approaches. Furthermore, it achieves a performance comparable to significantly larger models like Qwen3-32B, demonstrating superior effectiveness and robustness on open-domain tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2511_07922
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SERL: Self-Examining Reinforcement Learning on Open-Domain
Ou, Weixuan
Zheng, Yanzhao
Sun, Shuoshuo
Zhang, Wei
Dong, Baohua
Zhu, Hangcheng
Huang, Ruohui
Yu, Gang
Yan, Pengwei
Qiao, Yifan
Machine Learning
Reinforcement Learning (RL) has been shown to improve the capabilities of large language models (LLMs). However, applying RL to open-domain tasks faces two key challenges: (1) the inherent subjectivity of these tasks prevents the verifiable rewards as required by Reinforcement Learning with Verifiable Rewards (RLVR); (2) Reinforcement Learning from Human Feedback (RLHF) relies on external reward mechanisms. To overcome these limitations, we propose Self-Examining Reinforcement Learning (SERL), a novel self-improving framework where the LLM serves as both Actor and Judge. SERL introduces two synergistic reward mechanisms without any external signals. On the one hand, to improve the Actor's capability, we derive rewards from Copeland-style pairwise comparison judgments across a group of generated responses. On the other hand, a self-consistency reward that encourages coherent judgments is proposed to improve the Judge's reliability. This process refines the Judge's capability, which in turn provides a more robust reward for Actor. Experiments show that our method outperforms existing self-improvement training methods. SERL improves the LC win rate of Qwen3-8B on AlpacaEval 2 from 52.37% to 59.90%. To the best of our knowledge, our method achieves state-of-the-art performance among self-improving approaches. Furthermore, it achieves a performance comparable to significantly larger models like Qwen3-32B, demonstrating superior effectiveness and robustness on open-domain tasks.
title SERL: Self-Examining Reinforcement Learning on Open-Domain
topic Machine Learning
url https://arxiv.org/abs/2511.07922