Adapt to Thrive! Adaptive Power-Mean Policy Optimization for Improved LLM Reasoning

TL;DR

This paper introduces Adaptive Power-Mean Policy Optimization (APMPO), a novel reinforcement learning method that dynamically improves large language models' reasoning by adaptively tuning policy objectives and reward clipping.

Contribution

The paper proposes APMPO, combining Power-Mean Policy Optimization and Feedback-Adaptive Clipping, to enhance LLM reasoning and learning dynamics over static approaches.

Findings

APMPO outperforms state-of-the-art RLVR baselines on nine reasoning datasets.

Boosts Pass@1 score by 3.0 points on mathematical reasoning benchmarks.

Adaptive mechanisms improve model training stability and reasoning accuracy.

Abstract

Reinforcement Learning with Verifiable Rewards (RLVR) is an essential paradigm that enhances the reasoning capabilities of Large Language Models (LLMs). However, existing methods typically rely on static policy optimization schemes that misalign with the model's evolving reasoning capabilities. To address this issue, we propose Adaptive Power-Mean Policy Optimization (APMPO), which comprises two main innovations: Power-Mean Policy Optimization (PMPO) and Feedback-Adaptive Clipping (FAC). Specifically, PMPO introduces a generalized power-mean objective. This enables the model to adaptively transition from the signal-amplifying behavior of the arithmetic mean to the consistency-enforcing behavior of the geometric mean. FAC adaptively adjusts clipping bounds based on real-time reward statistics to overcome the limitations of static mechanisms. Capitalizing on these innovations, APMPO improves learning dynamics and reasoning performance. Extensive experiments on nine datasets across three reasoning tasks showcase the superiority of APMPO over state-of-the-art RLVR-based baselines. For instance, APMPO boosts the average Pass@1 score on mathematical reasoning benchmarks by 3.0 points compared to GRPO when using Qwen2.5-3B-Instruct.