Online Difficulty Filtering for Reasoning Oriented Reinforcement Learning

Fuente: arXiv
Enregistré dans:
Détails bibliographiques
Auteurs principaux: Bae, Sanghwan, Hong, Jiwoo, Lee, Min Young, Kim, Hanbyul, Nam, JeongYeon, Kwak, Donghyun
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866917219853664256
author Bae, Sanghwan
Hong, Jiwoo
Lee, Min Young
Kim, Hanbyul
Nam, JeongYeon
Kwak, Donghyun
author_facet Bae, Sanghwan
Hong, Jiwoo
Lee, Min Young
Kim, Hanbyul
Nam, JeongYeon
Kwak, Donghyun
contents Recent advances in reinforcement learning with verifiable rewards (RLVR) show that large language models enhance their reasoning abilities when trained with verifiable signals. However, due to reward sparsity, effectiveness depends heavily on selecting samples of appropriate difficulty. In this work, we present a formal analysis of online difficulty-aware filtering and establish its theoretical foundations. We show that expected policy improvement is lower-bounded by the variance of task-level success probabilities, implying that selecting tasks of intermediate difficulty maximizes learning efficiency. Building on this, we demonstrate that balanced filtering maximizes this lower bound, leading to superior performance and sample efficiency. Evaluations across multiple math reasoning benchmarks validate that balanced filtering consistently enhances convergence speed and final performance, achieving up to +12% gains in less than half the training steps of standard GRPO. By extending our analysis to various reward distributions, we provide a principled foundation for future RLVR curriculum strategies, confirmed through both theoretical analysis and extensive empirical results.
format Preprint
id arxiv_https___arxiv_org_abs_2504_03380
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Online Difficulty Filtering for Reasoning Oriented Reinforcement Learning
Bae, Sanghwan
Hong, Jiwoo
Lee, Min Young
Kim, Hanbyul
Nam, JeongYeon
Kwak, Donghyun
Computation and Language
Artificial Intelligence
Recent advances in reinforcement learning with verifiable rewards (RLVR) show that large language models enhance their reasoning abilities when trained with verifiable signals. However, due to reward sparsity, effectiveness depends heavily on selecting samples of appropriate difficulty. In this work, we present a formal analysis of online difficulty-aware filtering and establish its theoretical foundations. We show that expected policy improvement is lower-bounded by the variance of task-level success probabilities, implying that selecting tasks of intermediate difficulty maximizes learning efficiency. Building on this, we demonstrate that balanced filtering maximizes this lower bound, leading to superior performance and sample efficiency. Evaluations across multiple math reasoning benchmarks validate that balanced filtering consistently enhances convergence speed and final performance, achieving up to +12% gains in less than half the training steps of standard GRPO. By extending our analysis to various reward distributions, we provide a principled foundation for future RLVR curriculum strategies, confirmed through both theoretical analysis and extensive empirical results.
title Online Difficulty Filtering for Reasoning Oriented Reinforcement Learning
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2504.03380