ICLR 2026 Orals

Reducing Belief Deviation in Reinforcement Learning for Active Reasoning of LLM Agents

Deyu Zou, Yongqiang Chen, Jianxiang Wang, Garry YANG, Mufei Li, Qing Da, James Cheng, Pan Li, Yu Gong

LLMs & Reasoning Thu, Apr 23 · 10:42 AM–10:52 AM · Amphitheater Avg rating: 6.50 (4–8)

Abstract

Active reasoning requires large language model (LLM) agents to interact with external sources and strategically gather information to solve problems in multiple turns. Central to this process is belief tracking: maintaining an accurate representation of the underlying state and uncertainty in understanding and solving the problem. However, due to limited reasoning capabilities, LLM-based agents often suffer belief deviation: their internal beliefs drift from the true problem state, leading to loss of state awareness and uninformative or repetitive actions. Once this happens, errors compound in the trajectories used for reinforcement learning (RL), leading to misattributed credits and limited exploration. To address this issue, we propose to track belief deviation and develop $\mathbf{T^3}$, a simple yet principled method that detects excessive deviation and truncates training trajectories to suppress uninformative tail effects. Hence, $\mathbf{T^3}$ preserves credits for informative prefixes and systematically improves policy optimization. Across 5 challenging tasks, $\mathbf{T^3}$ consistently enhances training stability and yields performance gains of up to 30 points while cutting token cost by up to 34%. These results highlight belief control as a key principle for building robust LLM agents capable of active reasoning.

One-sentence summary·Auto-generated by claude-haiku-4-5-20251001(?)

Proposes T3 algorithm to detect belief deviation in LLM agents and truncate trajectories for improved reinforcement learning in active reasoning tasks.

Contributions·Auto-generated by claude-haiku-4-5-20251001(?)
  • Identifies belief deviation as critical failure mode in LLM-based active reasoning
  • T3 method that detects excessive deviation and truncates training trajectories
  • Achieves up to 30 point performance gains while reducing token cost by 34%
Methods used·Auto-generated by claude-haiku-4-5-20251001(?)
  • Reinforcement learning
  • Trajectory truncation
  • Belief tracking
  • Credit assignment
Limitations (author-stated)·Auto-generated by claude-haiku-4-5-20251001(?)

Authors did not state explicit limitations.

Future work (author-stated)·Auto-generated by claude-haiku-4-5-20251001(?)

Authors did not state explicit future directions.

Author keywords

  • Large language models
  • LLM reasoning
  • Agentic multi-turn reasoning

Related orals

Something off? Let us know →