Granular Concept Circuits: Toward a Fine-Grained Circuit Discovery for Concept Representations

2025-08-03

AI Summary

Deep visual models exhibit highly distributed representations of visual concepts, making precise localization of concept-specific neural substrates challenging. Method: This paper introduces the first fine-grained circuit discovery method tailored to specific visual concepts. It iteratively constructs semantically consistent and structurally interpretable concept-related neural circuits by jointly modeling functional dependencies among neurons and semantic alignment—without requiring human annotations. The approach supports parallel discovery and automatic parsing of multiple concepts. Contribution/Results: Evaluated on diverse mainstream image classification models—including ResNet and ViT—the method significantly improves spatial localization accuracy and semantic fidelity at the concept level. It provides a scalable, verifiable, and granular analytical framework for decoding internal model representations, advancing interpretability beyond coarse attribution or post-hoc explanation.

Technology Category

• Computer Vision: Representation Learning for Vision
• Machine Learning: Representation Learning
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Deep vision models have achieved remarkable classification performance by leveraging a hierarchical architecture in which human-interpretable concepts emerge through the composition of individual neurons across layers. Given the distributed nature of representations, pinpointing where specific visual concepts are encoded within a model remains a crucial yet challenging task. In this paper, we introduce an effective circuit discovery method, called Granular Concept Circuit (GCC), in which each circuit represents a concept relevant to a given query. To construct each circuit, our method iteratively assesses inter-neuron connectivity, focusing on both functional dependencies and semantic alignment. By automatically discovering multiple circuits, each capturing specific concepts within that query, our approach offers a profound, concept-wise interpretation of models and is the first to identify circuits tied to specific visual concepts at a fine-grained level. We validate the versatility and effectiveness of GCCs across various deep image classification models.

Problem

• Identify where visual concepts are encoded in deep models
• Discover fine-grained circuits for specific visual concepts
• Assess inter-neuron connectivity for functional and semantic alignment

Innovation

• Iteratively assesses inter-neuron connectivity
• Focuses on functional dependencies and semantic alignment
• Identifies fine-grained circuits for visual concepts