A gradient-guiding technique for Transformer attention interpretation yields detailed feature maps and reveals imperceptible image class-rewriting attacks on Vision Transformers.
National artifi- cial intelligence research and development strategic plan
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.AI 1years
2026 1verdicts
UNVERDICTED 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
Transformer Interpretability from Perspective of Attention and Gradient
A gradient-guiding technique for Transformer attention interpretation yields detailed feature maps and reveals imperceptible image class-rewriting attacks on Vision Transformers.