Metraipolis › archive › dc_ef4e92ab6254
Improving Math Reasoning through Value-guided Informative Search — arXiv:2610.01080
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.01080
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-06T12:03:41.402Z
- SHA-256 of the archived text
e4e8aed188140ad1f8a4cf2085b8aff2491afe3f7a800369188733dc42493a28- Stable URL of this page
https://metraipolis.com/documento/dc_ef4e92ab6254
How to cite
Improving Math Reasoning through Value-guided Informative Search — arXiv:2610.01080 — arXiv (en), https://arxiv.org/abs/2610.01080 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ef4e92ab6254 (Metraipolis archive dc_ef4e92ab6254, sha256 e4e8aed188140ad1f8a4cf2085b8aff2491afe3f7a800369188733dc42493a28)
BibTeX
@misc{dcef4e92ab6254,
title = {Improving Math Reasoning through Value-guided Informative Search — arXiv:2610.01080},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.01080},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ef4e92ab6254; sha256 e4e8aed188140ad1f8a4cf2085b8aff2491afe3f7a800369188733dc42493a28},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.01080. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Reinforcement learning with verifiable rewards (RLVR) has substantially improved the mathematical reasoning capabilities of large language models. Recent work introduces search into RLVR rollouts to increase trajectory diversity, but diversity alone does not ensure that the search-induced rollout policy improves upon the current policy. To address this gap, we propose APIVIS, a training-time framework that adapts finite-budget Gumbel search to chunk-level mathematical reasoning. APIVIS combines direct and searched responses within each rollout group, allowing improvements found by search to produce informative relative rewards. It further applies selective supervision to search-improved tokens, preserving a learning signal when uniform group rewards render GRPO ineffective. We show that exact value-guided selection improves the expected verifier reward at each searched state and that this guarantee extends to the complete rollout policy, with a corresponding approximate guarantee under bounded value-estimation error. Experiments on widely recognized mathematical reasoning benchmarks and different model scales demonstrate substantial improvements over competitive search-based methods, validating the effectiveness of APIVIS. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-10-02 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2610.01080 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-06T10:21:27.911Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-10-02?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).