Advantage Shaping as Surrogate Reward Maximization: Unifying Pass@K Policy Gradients
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908903962312704 |
|---|---|
| author | Thrampoulidis, Christos Mahdavi, Sadegh Deng, Wenlong |
| author_facet | Thrampoulidis, Christos Mahdavi, Sadegh Deng, Wenlong |
| contents | This note reconciles two seemingly distinct approaches to policy gradient optimization for the Pass@K objective in reinforcement learning with verifiable rewards: (1) direct REINFORCE-style methods, and (2) advantage-shaping techniques that directly modify GRPO. We show that these are two sides of the same coin. By reverse-engineering existing advantage-shaping algorithms, we reveal that they implicitly optimize surrogate rewards. We specifically interpret practical "hard-example up-weighting" modifications to GRPO as reward-level regularization. Conversely, starting from surrogate reward objectives, we provide a simple recipe for deriving both existing and new advantage-shaping methods. This perspective provides a lens for RLVR policy gradient optimization beyond our original motivation of Pass@K. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_23049 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Advantage Shaping as Surrogate Reward Maximization: Unifying Pass@K Policy Gradients Thrampoulidis, Christos Mahdavi, Sadegh Deng, Wenlong Machine Learning Artificial Intelligence This note reconciles two seemingly distinct approaches to policy gradient optimization for the Pass@K objective in reinforcement learning with verifiable rewards: (1) direct REINFORCE-style methods, and (2) advantage-shaping techniques that directly modify GRPO. We show that these are two sides of the same coin. By reverse-engineering existing advantage-shaping algorithms, we reveal that they implicitly optimize surrogate rewards. We specifically interpret practical "hard-example up-weighting" modifications to GRPO as reward-level regularization. Conversely, starting from surrogate reward objectives, we provide a simple recipe for deriving both existing and new advantage-shaping methods. This perspective provides a lens for RLVR policy gradient optimization beyond our original motivation of Pass@K. |
| title | Advantage Shaping as Surrogate Reward Maximization: Unifying Pass@K Policy Gradients |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2510.23049 |