Making Mathematical Reasoning Adaptive

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Lai, Zhejian, Geng, Xiang, Wang, Zhijun, Bai, Yang, Li, Jiahuan, Weng, Rongxiang, Wang, Jingang, Cao, Xuezhi, Cai, Xunliang, Huang, Shujian
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866918159119810560
author Lai, Zhejian
Geng, Xiang
Wang, Zhijun
Bai, Yang
Li, Jiahuan
Weng, Rongxiang
Wang, Jingang
Cao, Xuezhi
Cai, Xunliang
Huang, Shujian
author_facet Lai, Zhejian
Geng, Xiang
Wang, Zhijun
Bai, Yang
Li, Jiahuan
Weng, Rongxiang
Wang, Jingang
Cao, Xuezhi
Cai, Xunliang
Huang, Shujian
contents Mathematical reasoning is a primary indicator of large language models (LLMs) intelligence. However, existing LLMs exhibit failures of robustness and generalization. This paper attributes these deficiencies to spurious reasoning, i.e., producing answers from superficial features. To address this challenge, we propose the AdaR framework to enable adaptive reasoning, wherein models rely on problem-solving logic to produce answers. AdaR synthesizes logically equivalent queries by varying variable values, and trains models with RLVR on these data to penalize spurious logic while encouraging adaptive logic. To improve data quality, we extract the problem-solving logic from the original query and generate the corresponding answer by code execution, then apply a sanity check. Experimental results demonstrate that AdaR improves robustness and generalization, achieving substantial improvement in mathematical reasoning while maintaining high data efficiency. Analysis indicates that data synthesis and RLVR function in a coordinated manner to enable adaptive reasoning in LLMs. Subsequent analyses derive key design insights into the effect of critical factors and the applicability to instruct LLMs. Our project is available at https://github.com/NJUNLP/AdaR.
format Preprint
id arxiv_https___arxiv_org_abs_2510_04617
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Making Mathematical Reasoning Adaptive
Lai, Zhejian
Geng, Xiang
Wang, Zhijun
Bai, Yang
Li, Jiahuan
Weng, Rongxiang
Wang, Jingang
Cao, Xuezhi
Cai, Xunliang
Huang, Shujian
Artificial Intelligence
Mathematical reasoning is a primary indicator of large language models (LLMs) intelligence. However, existing LLMs exhibit failures of robustness and generalization. This paper attributes these deficiencies to spurious reasoning, i.e., producing answers from superficial features. To address this challenge, we propose the AdaR framework to enable adaptive reasoning, wherein models rely on problem-solving logic to produce answers. AdaR synthesizes logically equivalent queries by varying variable values, and trains models with RLVR on these data to penalize spurious logic while encouraging adaptive logic. To improve data quality, we extract the problem-solving logic from the original query and generate the corresponding answer by code execution, then apply a sanity check. Experimental results demonstrate that AdaR improves robustness and generalization, achieving substantial improvement in mathematical reasoning while maintaining high data efficiency. Analysis indicates that data synthesis and RLVR function in a coordinated manner to enable adaptive reasoning in LLMs. Subsequent analyses derive key design insights into the effect of critical factors and the applicability to instruct LLMs. Our project is available at https://github.com/NJUNLP/AdaR.
title Making Mathematical Reasoning Adaptive
topic Artificial Intelligence
url https://arxiv.org/abs/2510.04617