Research

Paper

AI LLM February 25, 2026

Enhancing Multi-Modal LLMs Reasoning via Difficulty-Aware Group Normalization

Authors

Jinghan Li, Junfeng Fang, Jinda Lu, Yuan Wang, Xiaoyan Guo, Tianyu Zhang, Xiang Wang, Xiangnan He

Abstract

Reinforcement Learning with Verifiable Rewards (RLVR) and Group Relative Policy Optimization (GRPO) have significantly advanced the reasoning capabilities of large language models. Extending these methods to multimodal settings, however, faces a critical challenge: the instability of std-based normalization, which is easily distorted by extreme samples with nearly positive or negative rewards. Unlike pure-text LLMs, multimodal models are particularly sensitive to such distortions, as both perceptual and reasoning errors influence their responses. To address this, we characterize each sample by its difficulty, defined through perceptual complexity (measured via visual entropy) and reasoning uncertainty (captured by model confidence). Building on this characterization, we propose difficulty-aware group normalization (Durian), which re-groups samples by difficulty levels and shares the std within each group. Our approach preserves GRPO's intra-group distinctions while eliminating sensitivity to extreme cases, yielding significant performance gains across multiple multimodal reasoning benchmarks.

Metadata

arXiv ID: 2602.21743
Provider: ARXIV
Primary Category: cs.CV
Published: 2026-02-25
Fetched: 2026-02-26 05:00

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2602.21743v1</id>\n    <title>Enhancing Multi-Modal LLMs Reasoning via Difficulty-Aware Group Normalization</title>\n    <updated>2026-02-25T09:52:50Z</updated>\n    <link href='https://arxiv.org/abs/2602.21743v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2602.21743v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Reinforcement Learning with Verifiable Rewards (RLVR) and Group Relative Policy Optimization (GRPO) have significantly advanced the reasoning capabilities of large language models. Extending these methods to multimodal settings, however, faces a critical challenge: the instability of std-based normalization, which is easily distorted by extreme samples with nearly positive or negative rewards. Unlike pure-text LLMs, multimodal models are particularly sensitive to such distortions, as both perceptual and reasoning errors influence their responses. To address this, we characterize each sample by its difficulty, defined through perceptual complexity (measured via visual entropy) and reasoning uncertainty (captured by model confidence). Building on this characterization, we propose difficulty-aware group normalization (Durian), which re-groups samples by difficulty levels and shares the std within each group. Our approach preserves GRPO's intra-group distinctions while eliminating sensitivity to extreme cases, yielding significant performance gains across multiple multimodal reasoning benchmarks.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CV'/>\n    <published>2026-02-25T09:52:50Z</published>\n    <arxiv:primary_category term='cs.CV'/>\n    <author>\n      <name>Jinghan Li</name>\n    </author>\n    <author>\n      <name>Junfeng Fang</name>\n    </author>\n    <author>\n      <name>Jinda Lu</name>\n    </author>\n    <author>\n      <name>Yuan Wang</name>\n    </author>\n    <author>\n      <name>Xiaoyan Guo</name>\n    </author>\n    <author>\n      <name>Tianyu Zhang</name>\n    </author>\n    <author>\n      <name>Xiang Wang</name>\n    </author>\n    <author>\n      <name>Xiangnan He</name>\n    </author>\n  </entry>"
}