Conditional Advantage Estimation for Reinforcement Learning in Large Reasoning Models

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Chen, Guanxu, Li, Yafu, Jiang, Yuxian, Qian, Chen, Ren, Qihan, Yang, Jingyi, Cheng, Yu, Liu, Dongrui, Shao, Jing
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866916974904213504
author Chen, Guanxu
Li, Yafu
Jiang, Yuxian
Qian, Chen
Ren, Qihan
Yang, Jingyi
Cheng, Yu
Liu, Dongrui
Shao, Jing
author_facet Chen, Guanxu
Li, Yafu
Jiang, Yuxian
Qian, Chen
Ren, Qihan
Yang, Jingyi
Cheng, Yu
Liu, Dongrui
Shao, Jing
contents Reinforcement Learning with Verifiable Rewards (RLVR) for large language models (LLMs) has achieved remarkable progress in enhancing LLMs' reasoning capabilities on tasks with clear correctness criteria, such as mathematical reasoning tasks. Several training metrics, such as entropy or response length, have been observed to correlate with different reasoning behaviors in reinforcement learning. Prior approaches incorporate such priors through reward or advantage shaping, which often relies on hand-crafted penalties and preferences (e.g., higher-is-better or lower-is-better). However, without careful hyperparameter tuning, these directional priors can be overly biased and may lead to failure. To this end, we introduce Conditional advANtage estimatiON (CANON), amplifying the impact of the target metric without presuming its direction. Specifically, CANON regroups the sampled responses into two groups based on the higher or lower value of a target metric, measures which metric trend contributes to better performance through inter-group comparison, and identifies the better response within the same group. In summary, CANON based on entropy consistently outperforms prior methods across three LLMs on both math reasoning and high-complexity logic tasks. When applied to response length, CANON further improves token efficiency, yielding a more favorable Pareto frontier in the performance-cost trade-off.
format Preprint
id arxiv_https___arxiv_org_abs_2509_23962
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Conditional Advantage Estimation for Reinforcement Learning in Large Reasoning Models
Chen, Guanxu
Li, Yafu
Jiang, Yuxian
Qian, Chen
Ren, Qihan
Yang, Jingyi
Cheng, Yu
Liu, Dongrui
Shao, Jing
Artificial Intelligence
Computation and Language
Reinforcement Learning with Verifiable Rewards (RLVR) for large language models (LLMs) has achieved remarkable progress in enhancing LLMs' reasoning capabilities on tasks with clear correctness criteria, such as mathematical reasoning tasks. Several training metrics, such as entropy or response length, have been observed to correlate with different reasoning behaviors in reinforcement learning. Prior approaches incorporate such priors through reward or advantage shaping, which often relies on hand-crafted penalties and preferences (e.g., higher-is-better or lower-is-better). However, without careful hyperparameter tuning, these directional priors can be overly biased and may lead to failure. To this end, we introduce Conditional advANtage estimatiON (CANON), amplifying the impact of the target metric without presuming its direction. Specifically, CANON regroups the sampled responses into two groups based on the higher or lower value of a target metric, measures which metric trend contributes to better performance through inter-group comparison, and identifies the better response within the same group. In summary, CANON based on entropy consistently outperforms prior methods across three LLMs on both math reasoning and high-complexity logic tasks. When applied to response length, CANON further improves token efficiency, yielding a more favorable Pareto frontier in the performance-cost trade-off.
title Conditional Advantage Estimation for Reinforcement Learning in Large Reasoning Models
topic Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2509.23962