AI RESEARCH
From Weights to Concepts: Data-Free Interpretability of CLIP via Singular Vector Decomposition
arXiv CS.CV
•
ArXi:2603.24653v1 Announce Type: new As vision-language models are deployed at scale, understanding their internal mechanisms becomes increasingly critical. Existing interpretability methods predominantly rely on activations, making them dataset-dependent, vulnerable to data bias, and often restricted to coarse head-level explanations. We