Abstract:Existing Reinforcement Learning from Verifiable Rewards (RLVR) methods, such as Group Relative Policy Optimization (GRPO), have achieved remarkable progress in improving the reasoning capabilities of Large Reasoning Models (LRMs). However, they exhibit limited exploration due to reliance on on-policy rollouts which are confined to the current policy's distribution, resulting in narrow trajectory diversity. Recent approaches attempt to expand policy coverage by incorporating trajectories generated from stronger expert models, yet this reliance increases computational cost and such advanced models are often inaccessible. To address these issues, we propose In-Context Steered Policy Optimization (ICPO), a unified framework that leverages the inherent in-context learning capability of LRMs to provide expert guidance using existing datasets. ICPO introduces mixed-policy GRPO with implicit expert forcing, which expands exploration beyond the current policy distribution without requiring advanced LRM trajectories. To further stabilize optimization, ICPO integrates expert region reject sampling to filter unreliable off-policy trajectories and annealed expert-bonus reward shaping to balance early expert guidance with later autonomous improvement. Results demonstrate that ICPO consistently enhances RLVR performance and training stability on mathematical reasoning benchmarks, revealing a scalable and effective RLVR paradigm for LRMs. Our code is available at this https URL.
| Comments: | ACL 2026 Findings. 19 pages, 13 figures, 12 tables |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2510.26519 [cs.LG] |
| (or arXiv:2510.26519v3 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2510.26519 arXiv-issued DOI via DataCite |
From: Hsiu-Yuan Huang [view email]
[v1]
Thu, 30 Oct 2025 14:14:15 UTC (973 KB)
[v2]
Wed, 7 Jan 2026 03:04:54 UTC (1,148 KB)
[v3]
Wed, 15 Apr 2026 11:36:27 UTC (1,069 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。