Visual Concept Connectome (VCC): Open World Concept Discovery and their Interlayer Connections in Deep Models

arXiv:2404.02233v1 Announce Type: new
Abstract: Understanding what deep network models capture in their learned representations is a fundamental challenge in computer vision. We present a new methodology to understanding such vision models, the Visual Concept Connectome (VCC), which discovers human interpretable concepts and their interlayer connections in a fully unsupervised manner. Our approach simultaneously reveals fine-grained concepts at a layer, connection weightings across all layers and is amendable to global analysis of network structure (e.g., branching pattern of hierarchical concept …

abstract arxiv challenge computer computer vision concept concepts cs.cv discovery human methodology network type understanding vision vision models visual world