Learning from Failures: Correction-Oriented Policy Optimization with Verifiable Rewards
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917494556459008 |
|---|---|
| author | Ren, Mengjie Lou, Jie Cao, Boxi Wen, Xueru Lin, Hongyu Han, Xianpei Sun, Le Yu, Xing Lu, Yaojie |
| author_facet | Ren, Mengjie Lou, Jie Cao, Boxi Wen, Xueru Lin, Hongyu Han, Xianpei Sun, Le Yu, Xing Lu, Yaojie |
| contents | Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as an effective paradigm for improving the reasoning capabilities of large language models. However, RLVR training is often hindered by sparse binary rewards and weak credit assignment, resulting in ambiguous optimization signals and underutilization of the useful information embedded in failed trajectories. To address this challenge, we propose Correction-Oriented Policy Optimization (CIPO), a simple and effective extension to RLVR that converts on-policy failed trajectories into correction-oriented supervision, without relying on any external signals. By jointly optimizing correction samples derived from the model's own failed attempts together with the standard RLVR objective, CIPO improves learning effectiveness while explicitly enhancing the model's ability to correct its own errors. Extensive experiments across 11 benchmarks spanning mathematical reasoning and code generation demonstrate that CIPO consistently and significantly outperforms strong baselines in both reasoning and correction performance. Moreover, CIPO yields stronger pass@K gains, indicating that it improves the model's intrinsic reasoning capacity rather than merely redistributing probability mass over existing correct answers. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_14539 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Learning from Failures: Correction-Oriented Policy Optimization with Verifiable Rewards Ren, Mengjie Lou, Jie Cao, Boxi Wen, Xueru Lin, Hongyu Han, Xianpei Sun, Le Yu, Xing Lu, Yaojie Computation and Language Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as an effective paradigm for improving the reasoning capabilities of large language models. However, RLVR training is often hindered by sparse binary rewards and weak credit assignment, resulting in ambiguous optimization signals and underutilization of the useful information embedded in failed trajectories. To address this challenge, we propose Correction-Oriented Policy Optimization (CIPO), a simple and effective extension to RLVR that converts on-policy failed trajectories into correction-oriented supervision, without relying on any external signals. By jointly optimizing correction samples derived from the model's own failed attempts together with the standard RLVR objective, CIPO improves learning effectiveness while explicitly enhancing the model's ability to correct its own errors. Extensive experiments across 11 benchmarks spanning mathematical reasoning and code generation demonstrate that CIPO consistently and significantly outperforms strong baselines in both reasoning and correction performance. Moreover, CIPO yields stronger pass@K gains, indicating that it improves the model's intrinsic reasoning capacity rather than merely redistributing probability mass over existing correct answers. |
| title | Learning from Failures: Correction-Oriented Policy Optimization with Verifiable Rewards |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2605.14539 |