Circuit Tracing in Vision-Language Models: Understanding the Internal Mechanisms of Multimodal Thinking

2026-02-23

AI Summary

This work addresses the opacity of vision-language models (VLMs) as black-box systems by proposing the first transparent circuit tracing framework tailored for VLMs. Leveraging a transcoder, attribution graphs, and attention analysis, the framework systematically identifies and causally validates the role of visual feature circuits in semantic integration. For the first time, it enables causal intervention and repair of internal vision-to-semantic pathways within VLMs, uncovering their critical functions in mathematical reasoning and cross-modal association. Experimental results demonstrate that the identified circuits exhibit strong causal efficacy and controllability, substantially enhancing model interpretability and reliability. This advancement lays a foundational step toward building trustworthy multimodal systems.

Technology Category

• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Vision-language models (VLMs) are powerful but remain opaque black boxes. We introduce the first framework for transparent circuit tracing in VLMs to systematically analyze multimodal reasoning. By utilizing transcoders, attribution graphs, and attention-based methods, we uncover how VLMs hierarchically integrate visual and semantic concepts. We reveal that distinct visual feature circuits can handle mathematical reasoning and support cross-modal associations. Validated through feature steering and circuit patching, our framework proves these circuits are causal and controllable, laying the groundwork for more explainable and reliable VLMs.

Problem

• vision-language models
• circuit tracing
• multimodal reasoning
• model interpretability
• cross-modal association

Innovation

• circuit tracing
• vision-language models
• multimodal reasoning
• attribution graphs
• feature steering