Enhancing Multi-Modal LLMs Reasoning via Difficulty-Aware Group Normalization
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866918357067890688 |
|---|---|
| author | Li, Jinghan Fang, Junfeng Lu, Jinda Wang, Yuan Guo, Xiaoyan Zhang, Tianyu Wang, Xiang He, Xiangnan |
| author_facet | Li, Jinghan Fang, Junfeng Lu, Jinda Wang, Yuan Guo, Xiaoyan Zhang, Tianyu Wang, Xiang He, Xiangnan |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_21743 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Enhancing Multi-Modal LLMs Reasoning via Difficulty-Aware Group Normalization Li, Jinghan Fang, Junfeng Lu, Jinda Wang, Yuan Guo, Xiaoyan Zhang, Tianyu Wang, Xiang He, Xiangnan Computer Vision and Pattern Recognition 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. |
| title | Enhancing Multi-Modal LLMs Reasoning via Difficulty-Aware Group Normalization |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2602.21743 |