Signals 4 · free daily AI digest

"World Knowledge" in the Weights: Reading Concept Circuits of Vision Transformers

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

Abstract

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

Read on arXiv →

#115 most recent of 237 cs.CV papers we have recorded · ↑ newer: Rethinking Learned Occupancy in Autonomous Active Mapping with Observa · ↓ older: Task-driven Processing with Coarse-to-Fine Glimpse-based Active Percep
Cite this page: "World Knowledge" in the Weights: Reading Concept Circuits of Vision Transformers: the #115 most recent of 237 cs.CV papers we have recorded (as of 2026-09-08). Source: Signals 4 (Signals API) — https://data.jiangzhang.ca/signals4/t/papers/world-knowledge-in-the-weights-reading-concept-circuits-of-vision-transformers.html
Free to quote with attribution to “Signals 4 (Signals API)”. Machine-readable: papers.json
Related: More cs.CV papers · arXiv signals · All papers · Today in AI
Get 4 AI signals a day by email — free.
Subscribe free → See all plans →
Get 4 AI signals a day by email — free
All models · All repos · By company · Daily editions