Enregistré dans:
Détails bibliographiques
Auteurs principaux: Zhang, Heng, Hu, Haichuan, Shen, Yaomin, Yu, Weihao, Yuan, Yilei, You, Haochen, Cheng, Guo, Zhang, Zijian, Gan, Lubin, Wei, Huihui, Zhang, Hao, Huang, Jin
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:https://arxiv.org/abs/2509.12715
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866914213432131584
author Zhang, Heng
Hu, Haichuan
Shen, Yaomin
Yu, Weihao
Yuan, Yilei
You, Haochen
Cheng, Guo
Zhang, Zijian
Gan, Lubin
Wei, Huihui
Zhang, Hao
Huang, Jin
author_facet Zhang, Heng
Hu, Haichuan
Shen, Yaomin
Yu, Weihao
Yuan, Yilei
You, Haochen
Cheng, Guo
Zhang, Zijian
Gan, Lubin
Wei, Huihui
Zhang, Hao
Huang, Jin
contents Large Vision-Language Models (LVLMs) have demonstrated impressive performance on multimodal tasks through scaled architectures and extensive training. However, existing Mixture of Experts (MoE) approaches face challenges due to the asymmetry between visual and linguistic processing. Visual information is spatially complete, while language requires maintaining sequential context. As a result, MoE models struggle to balance modality-specific features and cross-modal interactions. Through systematic analysis, we observe that language experts in deeper layers progressively lose contextual grounding and rely more on parametric knowledge rather than utilizing the provided visual and linguistic information. To address this, we propose AsyMoE, a novel architecture that models this asymmetry using three specialized expert groups. We design intra-modality experts for modality-specific processing, hyperbolic inter-modality experts for hierarchical cross-modal interactions, and evidence-priority language experts to suppress parametric biases and maintain contextual grounding. Extensive experiments demonstrate that AsyMoE achieves 26.58% and 15.45% accuracy improvements over vanilla MoE and modality-specific MoE respectively, with 25.45% fewer activated parameters than dense models.
format Preprint
id arxiv_https___arxiv_org_abs_2509_12715
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle AsyMoE: Leveraging Modal Asymmetry for Enhanced Expert Specialization in Large Vision-Language Models
Zhang, Heng
Hu, Haichuan
Shen, Yaomin
Yu, Weihao
Yuan, Yilei
You, Haochen
Cheng, Guo
Zhang, Zijian
Gan, Lubin
Wei, Huihui
Zhang, Hao
Huang, Jin
Computer Vision and Pattern Recognition
Robotics
Large Vision-Language Models (LVLMs) have demonstrated impressive performance on multimodal tasks through scaled architectures and extensive training. However, existing Mixture of Experts (MoE) approaches face challenges due to the asymmetry between visual and linguistic processing. Visual information is spatially complete, while language requires maintaining sequential context. As a result, MoE models struggle to balance modality-specific features and cross-modal interactions. Through systematic analysis, we observe that language experts in deeper layers progressively lose contextual grounding and rely more on parametric knowledge rather than utilizing the provided visual and linguistic information. To address this, we propose AsyMoE, a novel architecture that models this asymmetry using three specialized expert groups. We design intra-modality experts for modality-specific processing, hyperbolic inter-modality experts for hierarchical cross-modal interactions, and evidence-priority language experts to suppress parametric biases and maintain contextual grounding. Extensive experiments demonstrate that AsyMoE achieves 26.58% and 15.45% accuracy improvements over vanilla MoE and modality-specific MoE respectively, with 25.45% fewer activated parameters than dense models.
title AsyMoE: Leveraging Modal Asymmetry for Enhanced Expert Specialization in Large Vision-Language Models
topic Computer Vision and Pattern Recognition
Robotics
url https://arxiv.org/abs/2509.12715