Training-free mechanistic interpretability locates lexical-encoding attention heads in ViT and shows that targeted interventions on them improve robustness to typographic attacks in CLIP and downstream LVLMs.
arXiv preprint arXiv:2506.17052 (2025)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Towards Robustness against Typographic Attack with Training-free Concept Localization
Training-free mechanistic interpretability locates lexical-encoding attention heads in ViT and shows that targeted interventions on them improve robustness to typographic attacks in CLIP and downstream LVLMs.