Saved in:
Bibliographic Details
Main Authors: Li, Zhaochun, Wang, Chen, Bai, Jionghao, Cui, Shisheng, Lan, Ge, Zhao, Zhou, Wang, Yue
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.12730
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917209653116928
author Li, Zhaochun
Wang, Chen
Bai, Jionghao
Cui, Shisheng
Lan, Ge
Zhao, Zhou
Wang, Yue
author_facet Li, Zhaochun
Wang, Chen
Bai, Jionghao
Cui, Shisheng
Lan, Ge
Zhao, Zhou
Wang, Yue
contents The exploration-exploitation (EE) trade-off is a central challenge in reinforcement learning (RL) for large language models (LLMs). With Group Relative Policy Optimization (GRPO), training tends to be exploitation driven: entropy decreases monotonically, samples convergence, and exploration fades. Most existing fixes are \textbf{sample-centric}: they seek or bonus rare samples, assuming exploration comes from novel trajectories and tokens. These heuristics depend on the "luck" of informative samples, lack principled control of the policy, and often yield limited or inconsistent gains. In this work, we are the first to introduce a \textbf{distribution-centric} perspective for RL, in which exploration is always guided by a "better" target distribution, and reveal that a policy's ability to resist entropy collapse is governed by the distribution itself rather than individual samples. Building on this insight, we propose Distribution-Centric Policy Optimization (DCPO), which reformulates entropy regulation as distribution-level regularization. DCPO achieves controllable entropy fully on-policy without sampling from external distributions, enabling efficient exploration while maintaining training stability. Across multiple models and seven benchmarks, DCPO improves over GRPO by about 20\% on average. Overall, DCPO replaces sample-level heuristics with distribution-level principles, offering a theoretically grounded and flexible framework for controllable exploration and a stronger EE trade-off. The code is available in https://github.com/597358816/DCPO.
format Preprint
id arxiv_https___arxiv_org_abs_2601_12730
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Distribution-Centric Policy Optimization Dominates Exploration-Exploitation Trade-off
Li, Zhaochun
Wang, Chen
Bai, Jionghao
Cui, Shisheng
Lan, Ge
Zhao, Zhou
Wang, Yue
Machine Learning
The exploration-exploitation (EE) trade-off is a central challenge in reinforcement learning (RL) for large language models (LLMs). With Group Relative Policy Optimization (GRPO), training tends to be exploitation driven: entropy decreases monotonically, samples convergence, and exploration fades. Most existing fixes are \textbf{sample-centric}: they seek or bonus rare samples, assuming exploration comes from novel trajectories and tokens. These heuristics depend on the "luck" of informative samples, lack principled control of the policy, and often yield limited or inconsistent gains. In this work, we are the first to introduce a \textbf{distribution-centric} perspective for RL, in which exploration is always guided by a "better" target distribution, and reveal that a policy's ability to resist entropy collapse is governed by the distribution itself rather than individual samples. Building on this insight, we propose Distribution-Centric Policy Optimization (DCPO), which reformulates entropy regulation as distribution-level regularization. DCPO achieves controllable entropy fully on-policy without sampling from external distributions, enabling efficient exploration while maintaining training stability. Across multiple models and seven benchmarks, DCPO improves over GRPO by about 20\% on average. Overall, DCPO replaces sample-level heuristics with distribution-level principles, offering a theoretically grounded and flexible framework for controllable exploration and a stronger EE trade-off. The code is available in https://github.com/597358816/DCPO.
title Distribution-Centric Policy Optimization Dominates Exploration-Exploitation Trade-off
topic Machine Learning
url https://arxiv.org/abs/2601.12730