Saved in:
Bibliographic Details
Main Authors: Xiu, Lixin, Luo, Xufang, Nakayama, Hideki
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.29676
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911557759270912
author Xiu, Lixin
Luo, Xufang
Nakayama, Hideki
author_facet Xiu, Lixin
Luo, Xufang
Nakayama, Hideki
contents Large vision-language models (LVLMs) achieve impressive performance, yet their internal decision-making processes remain opaque, making it difficult to determine if the success stems from true multimodal fusion or from reliance on unimodal priors. To address this attribution gap, we introduce a novel framework using partial information decomposition (PID) to quantitatively measure the "information spectrum" of LVLMs -- decomposing a model's decision-relevant information into redundant, unique, and synergistic components. By adapting a scalable estimator to modern LVLM outputs, our model-agnostic pipeline profiles 26 LVLMs on four datasets across three dimensions -- breadth (cross-model & cross-task), depth (layer-wise information dynamics), and time (learning dynamics across training). Our analysis reveals two key results: (i) two task regimes (synergy-driven vs. knowledge-driven) and (ii) two stable, contrasting family-level strategies (fusion-centric vs. language-centric). We also uncover a consistent three-phase pattern in layer-wise processing and identify visual instruction tuning as the key stage where fusion is learned. Together, these contributions provide a quantitative lens beyond accuracy-only evaluation and offer insights for analyzing and designing the next generation of LVLMs. Code and data are available at https://github.com/RiiShin/pid-lvlm-analysis .
format Preprint
id arxiv_https___arxiv_org_abs_2603_29676
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle A Comprehensive Information-Decomposition Analysis of Large Vision-Language Models
Xiu, Lixin
Luo, Xufang
Nakayama, Hideki
Machine Learning
Computation and Language
Computer Vision and Pattern Recognition
Large vision-language models (LVLMs) achieve impressive performance, yet their internal decision-making processes remain opaque, making it difficult to determine if the success stems from true multimodal fusion or from reliance on unimodal priors. To address this attribution gap, we introduce a novel framework using partial information decomposition (PID) to quantitatively measure the "information spectrum" of LVLMs -- decomposing a model's decision-relevant information into redundant, unique, and synergistic components. By adapting a scalable estimator to modern LVLM outputs, our model-agnostic pipeline profiles 26 LVLMs on four datasets across three dimensions -- breadth (cross-model & cross-task), depth (layer-wise information dynamics), and time (learning dynamics across training). Our analysis reveals two key results: (i) two task regimes (synergy-driven vs. knowledge-driven) and (ii) two stable, contrasting family-level strategies (fusion-centric vs. language-centric). We also uncover a consistent three-phase pattern in layer-wise processing and identify visual instruction tuning as the key stage where fusion is learned. Together, these contributions provide a quantitative lens beyond accuracy-only evaluation and offer insights for analyzing and designing the next generation of LVLMs. Code and data are available at https://github.com/RiiShin/pid-lvlm-analysis .
title A Comprehensive Information-Decomposition Analysis of Large Vision-Language Models
topic Machine Learning
Computation and Language
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2603.29676