Paper recorded by Signals 4 on 2026-09-08 in cs.CV. Abstract reproduced from arXiv; link to the original below.
Published 2026-09-08 on arXiv · recorded by Signals 4 on 2026-09-09
Category: cs.CV · 计算机视觉 · first seen 2026-09-09
Vision transformers (ViTs) have achieved remarkable generalization across visual domains, yet little is known about how they internally represent the structure of the world. To address this gap, we use Cross-Layer Transcoders (CLTs) to read concept circuits from ViTs: directed graphs whose nodes correspond to sparse, interpretable concepts and edges capture concept interactions across layers. Our