Latent Poincaré Shaping for Agentic Reinforcement Learning
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_ | 1866918381753466880 |
|---|---|
| author | Xia, Hanchen Chen, Baoyou Zang, Zelin Ge, Yutang Zhao, Guojiang Zhu, Siyu |
| author_facet | Xia, Hanchen Chen, Baoyou Zang, Zelin Ge, Yutang Zhao, Guojiang Zhu, Siyu |
| contents | We propose LaPha, a method for training AlphaZero-like LLM agents in a Poincaré latent space. Under LaPha, the search process can be visualized as a tree rooted at the prompt and growing outward from the origin toward the boundary of the Poincaré ball, where negative curvature provides exponentially increasing capacity with radius. Using hyperbolic geodesic distance to rule-verified correctness, we define a node potential and assign dense process rewards by potential differences. We further attach a lightweight value head on the same shared latent space, enabling self-guided test-time scaling with almost no additional overhead. On MATH-500, LaPha improves Qwen2.5-Math-1.5B from 66.0% to 88.2%. With value-head-guided search, LaPha-1.5B reaches 56.7% accuracy on AIME'24, and LaPha-7B further achieves 60.0% on AIME'24 and 53.3% on AIME'25. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_09375 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Latent Poincaré Shaping for Agentic Reinforcement Learning Xia, Hanchen Chen, Baoyou Zang, Zelin Ge, Yutang Zhao, Guojiang Zhu, Siyu Machine Learning We propose LaPha, a method for training AlphaZero-like LLM agents in a Poincaré latent space. Under LaPha, the search process can be visualized as a tree rooted at the prompt and growing outward from the origin toward the boundary of the Poincaré ball, where negative curvature provides exponentially increasing capacity with radius. Using hyperbolic geodesic distance to rule-verified correctness, we define a node potential and assign dense process rewards by potential differences. We further attach a lightweight value head on the same shared latent space, enabling self-guided test-time scaling with almost no additional overhead. On MATH-500, LaPha improves Qwen2.5-Math-1.5B from 66.0% to 88.2%. With value-head-guided search, LaPha-1.5B reaches 56.7% accuracy on AIME'24, and LaPha-7B further achieves 60.0% on AIME'24 and 53.3% on AIME'25. |
| title | Latent Poincaré Shaping for Agentic Reinforcement Learning |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2602.09375 |