Saved in:
Bibliographic Details
Main Authors: Li, Xiaofan, Yang, Ming, Ma, Zhiyuan, Ma, Shichao, Du, Jintao, Cheng, Yu, Wang, Weiqiang, Zhang, Zhizhong, Tan, Xin, Qu, Yanyun, Ma, Lizhuang, Xie, Yuan
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.13902
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915938754887680
author Li, Xiaofan
Yang, Ming
Ma, Zhiyuan
Ma, Shichao
Du, Jintao
Cheng, Yu
Wang, Weiqiang
Zhang, Zhizhong
Tan, Xin
Qu, Yanyun
Ma, Lizhuang
Xie, Yuan
author_facet Li, Xiaofan
Yang, Ming
Ma, Zhiyuan
Ma, Shichao
Du, Jintao
Cheng, Yu
Wang, Weiqiang
Zhang, Zhizhong
Tan, Xin
Qu, Yanyun
Ma, Lizhuang
Xie, Yuan
contents Reinforcement Learning with Verifiable Rewards (RLVR) has catalyzed significant advances in the reasoning capabilities of Large Language Models (LLMs). However, effectively managing the exploration and exploitation trade-off remains a critical challenge. In this paper, we fully analyze the exploration and exploitation dilemma of extremely hard and easy samples during the training and propose a new fine-grained trade-off mechanism. Concretely, we introduce a perplexity space disentangling strategy that divides the sample space into distinct exploration (high perplexity) and exploitation (low perplexity) subspaces, thereby mining fine-grained samples requiring exploration-exploitation trade-off. Subsequently, we propose a bidirectional reward allocation mechanism with a minimum impact on verification rewards to implement perplexity-guided exploration and exploitation, enabling more stable policy optimization. Finally, we have evaluated our method on two mainstream tasks: mathematical reasoning and function calling, and experimental results demonstrate the superiority of the proposed method, confirming its effectiveness in enhancing LLM performance by fine-grained exploration-exploitation trade-off.
format Preprint
id arxiv_https___arxiv_org_abs_2604_13902
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle DiPO: Disentangled Perplexity Policy Optimization for Fine-grained Exploration-Exploitation Trade-Off
Li, Xiaofan
Yang, Ming
Ma, Zhiyuan
Ma, Shichao
Du, Jintao
Cheng, Yu
Wang, Weiqiang
Zhang, Zhizhong
Tan, Xin
Qu, Yanyun
Ma, Lizhuang
Xie, Yuan
Machine Learning
Reinforcement Learning with Verifiable Rewards (RLVR) has catalyzed significant advances in the reasoning capabilities of Large Language Models (LLMs). However, effectively managing the exploration and exploitation trade-off remains a critical challenge. In this paper, we fully analyze the exploration and exploitation dilemma of extremely hard and easy samples during the training and propose a new fine-grained trade-off mechanism. Concretely, we introduce a perplexity space disentangling strategy that divides the sample space into distinct exploration (high perplexity) and exploitation (low perplexity) subspaces, thereby mining fine-grained samples requiring exploration-exploitation trade-off. Subsequently, we propose a bidirectional reward allocation mechanism with a minimum impact on verification rewards to implement perplexity-guided exploration and exploitation, enabling more stable policy optimization. Finally, we have evaluated our method on two mainstream tasks: mathematical reasoning and function calling, and experimental results demonstrate the superiority of the proposed method, confirming its effectiveness in enhancing LLM performance by fine-grained exploration-exploitation trade-off.
title DiPO: Disentangled Perplexity Policy Optimization for Fine-grained Exploration-Exploitation Trade-Off
topic Machine Learning
url https://arxiv.org/abs/2604.13902