analiza aiqbench
This research introduces GeoPID, a training-free framework that geometrically decomposes multimodal information in vision-language models to better leverage visual inputs over textual context. By selectively amplifying vision-unique subspaces during inference, the method achieves a significant 7.63% accuracy gain on visual grounding tasks without updating model parameters. The work offers a novel analytical lens for understanding and steering representation spaces within large multimodal models.
GeoPID: Decomposing and Steering Visual Information in Vision-Language Models
arXiv:2610.08401v1 Announce Type: cross Abstract: While recent vision-language models (VLMs) have shown outstanding performance across diverse applications, they tend to under-use visual information and over-rely on textual context. In this work, we propose \textsc{GeoPID}, a training-free framework that analyzes multimodal information within VLMs from a geometric perspective. \textsc{GeoPID} decomposes information into Redundant, Modality-Unique, and Synergistic components through the geometric relationships between visual and textual representation subspaces. Through an extensive analysis across 22 VLMs and 14 benchmarks, we confirm that correct predictions exhibit stronger vision-unique components when questions strongly require visual grounding. Building on this geometric analysis, we introduce a targeted intervention technique that selectively amplifies visual representations along the vision-unique subspace during inference. As a result, visual grounding capabilities were enhanced without any additional model parameter updates, achieving an average relative accuracy gain of 7.63\%.
Abstract:While recent vision-language models (VLMs) have shown outstanding performance across diverse applications, they tend to under-use visual information and over-rely on textual context. In this work, we propose \textsc{GeoPID}, a training-free framework that analyzes multimodal information within VLMs from a geometric perspective. \textsc{GeoPID} decomposes information into Redundant, Modality-Unique, and Synergistic components through the geometric relationships between visual and textual representation subspaces. Through an extensive analysis across 22 VLMs and 14 benchmarks, we confirm that correct predictions exhibit stronger vision-unique components when questions strongly require visual grounding. Building on this geometric analysis, we introduce a targeted intervention technique that selectively amplifies visual representations along the vision-unique subspace during inference. As a result, visual grounding capabilities were enhanced without any additional model parameter updates, achieving an average relative accuracy gain of 7.63\%.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2610.08401 [cs.CV]
(or arXiv:2610.08401v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2610.08401 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 230 cuvinte
Subiecte conexe
Linguistic Context Recodes Visual Representations in Vision-Language Models
arXiv · 5 aug. 2026
What do VLM-Based Vision-Language Navigation Models Rely on: Interpreting and Steering Policy Behavior
arXiv · 22 sept. 2026
[AINews] Claude Opus 5.5, the new default model for AINews — and everybody cuts prices 40-50%
Latent Space · 23 sept. 2026
Describe-Then-Act: Proactive Agent Steering via Distilled Language-Action World Models
arXiv · 1 sept. 2026