Saved in:
Bibliographic Details
Main Authors: Wang, Jiawei, Liu, Jiacai, Fu, Yuqian, Li, Yingru, Wang, Xintao, Lin, Yuan, Yue, Yu, Zhang, Lin, Wang, Yang, Wang, Ke
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2509.09265
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912582151962624
author Wang, Jiawei
Liu, Jiacai
Fu, Yuqian
Li, Yingru
Wang, Xintao
Lin, Yuan
Yue, Yu
Zhang, Lin
Wang, Yang
Wang, Ke
author_facet Wang, Jiawei
Liu, Jiacai
Fu, Yuqian
Li, Yingru
Wang, Xintao
Lin, Yuan
Yue, Yu
Zhang, Lin
Wang, Yang
Wang, Ke
contents In long-horizon tasks, recent agents based on Large Language Models (LLMs) face a significant challenge that sparse, outcome-based rewards make it difficult to assign credit to intermediate steps. Previous methods mainly focus on creating dense reward signals to guide learning, either through traditional reinforcement learning techniques like inverse reinforcement learning or by using Process Reward Models for step-by-step feedback. In this paper, we identify a fundamental problem in the learning dynamics of LLMs: the magnitude of policy gradients is inherently coupled with the entropy, which leads to inefficient small updates for confident correct actions and potentially destabilizes large updates for uncertain ones. To resolve this, we propose Entropy-Modulated Policy Gradients (EMPG), a framework that re-calibrates the learning signal based on step-wise uncertainty and the final task outcome. EMPG amplifies updates for confident correct actions, penalizes confident errors, and attenuates updates from uncertain steps to stabilize exploration. We further introduce a bonus term for future clarity that encourages agents to find more predictable solution paths. Through comprehensive experiments on three challenging agent tasks, WebShop, ALFWorld, and Deep Search, we demonstrate that EMPG achieves substantial performance gains and significantly outperforms strong policy gradient baselines. Project page is at https://empgseed-seed.github.io/
format Preprint
id arxiv_https___arxiv_org_abs_2509_09265
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Harnessing Uncertainty: Entropy-Modulated Policy Gradients for Long-Horizon LLM Agents
Wang, Jiawei
Liu, Jiacai
Fu, Yuqian
Li, Yingru
Wang, Xintao
Lin, Yuan
Yue, Yu
Zhang, Lin
Wang, Yang
Wang, Ke
Machine Learning
Computation and Language
In long-horizon tasks, recent agents based on Large Language Models (LLMs) face a significant challenge that sparse, outcome-based rewards make it difficult to assign credit to intermediate steps. Previous methods mainly focus on creating dense reward signals to guide learning, either through traditional reinforcement learning techniques like inverse reinforcement learning or by using Process Reward Models for step-by-step feedback. In this paper, we identify a fundamental problem in the learning dynamics of LLMs: the magnitude of policy gradients is inherently coupled with the entropy, which leads to inefficient small updates for confident correct actions and potentially destabilizes large updates for uncertain ones. To resolve this, we propose Entropy-Modulated Policy Gradients (EMPG), a framework that re-calibrates the learning signal based on step-wise uncertainty and the final task outcome. EMPG amplifies updates for confident correct actions, penalizes confident errors, and attenuates updates from uncertain steps to stabilize exploration. We further introduce a bonus term for future clarity that encourages agents to find more predictable solution paths. Through comprehensive experiments on three challenging agent tasks, WebShop, ALFWorld, and Deep Search, we demonstrate that EMPG achieves substantial performance gains and significantly outperforms strong policy gradient baselines. Project page is at https://empgseed-seed.github.io/
title Harnessing Uncertainty: Entropy-Modulated Policy Gradients for Long-Horizon LLM Agents
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2509.09265