DEAL: Disentangle and Localize Concept-level Explanations for VLMs

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Li, Tang, Ma, Mengmeng, Peng, Xi
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914877674618880
author Li, Tang
Ma, Mengmeng
Peng, Xi
author_facet Li, Tang
Ma, Mengmeng
Peng, Xi
contents Large pre-trained Vision-Language Models (VLMs) have become ubiquitous foundational components of other models and downstream tasks. Although powerful, our empirical results reveal that such models might not be able to identify fine-grained concepts. Specifically, the explanations of VLMs with respect to fine-grained concepts are entangled and mislocalized. To address this issue, we propose to DisEntAngle and Localize (DEAL) the concept-level explanations for VLMs without human annotations. The key idea is encouraging the concept-level explanations to be distinct while maintaining consistency with category-level explanations. We conduct extensive experiments and ablation studies on a wide range of benchmark datasets and vision-language models. Our empirical results demonstrate that the proposed method significantly improves the concept-level explanations of the model in terms of disentanglability and localizability. Surprisingly, the improved explainability alleviates the model's reliance on spurious correlations, which further benefits the prediction accuracy.
format Preprint
id arxiv_https___arxiv_org_abs_2407_14412
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle DEAL: Disentangle and Localize Concept-level Explanations for VLMs
Li, Tang
Ma, Mengmeng
Peng, Xi
Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
Large pre-trained Vision-Language Models (VLMs) have become ubiquitous foundational components of other models and downstream tasks. Although powerful, our empirical results reveal that such models might not be able to identify fine-grained concepts. Specifically, the explanations of VLMs with respect to fine-grained concepts are entangled and mislocalized. To address this issue, we propose to DisEntAngle and Localize (DEAL) the concept-level explanations for VLMs without human annotations. The key idea is encouraging the concept-level explanations to be distinct while maintaining consistency with category-level explanations. We conduct extensive experiments and ablation studies on a wide range of benchmark datasets and vision-language models. Our empirical results demonstrate that the proposed method significantly improves the concept-level explanations of the model in terms of disentanglability and localizability. Surprisingly, the improved explainability alleviates the model's reliance on spurious correlations, which further benefits the prediction accuracy.
title DEAL: Disentangle and Localize Concept-level Explanations for VLMs
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2407.14412