A Survey on Mechanistic Interpretability for Multi-Modal Foundation Models
Abstract
The rise of foundation models has transformed machine learning research, prompting efforts to uncover their inner workings and develop more efficient and reliable applications for better control. While significant progress has been made in interpreting Large Language Models (LLMs), multimodal foundation models (MMFMs) - such as contrastive vision-language models, generative vision-language models, and text-to-image models - pose unique interpretability challenges beyond unimodal frameworks. Despite initial studies, a substantial gap remains between the interpretability of LLMs and MMFMs. This survey explores two key aspects: (1) the adaptation of LLM interpretability methods to multimodal models and (2) understanding the mechanistic differences between unimodal language models and crossmodal systems. By systematically reviewing current MMFM analysis techniques, we propose a structured taxonomy of interpretability methods, compare insights across unimodal and multimodal architectures, and highlight critical research gaps.
Cite
@article{arxiv.2502.17516,
title = {A Survey on Mechanistic Interpretability for Multi-Modal Foundation Models},
author = {Zihao Lin and Samyadeep Basu and Mohammad Beigi and Varun Manjunatha and Ryan A. Rossi and Zichao Wang and Yufan Zhou and Sriram Balasubramanian and Arman Zarei and Keivan Rezaei and Ying Shen and Barry Menglong Yao and Zhiyang Xu and Qin Liu and Yuxiang Zhang and Yan Sun and Shilong Liu and Li Shen and Hongxuan Li and Soheil Feizi and Lifu Huang},
journal= {arXiv preprint arXiv:2502.17516},
year = {2025}
}
Comments
30 pages, 4 Figures, 10 Tables