Aligning adversarial perturbations with the near-null singular directions of intermediate linear layers in transformer VLMs yields stronger attacks than existing feature- and output-space methods.
In: Forty-second International Conference on Machine Learning (2025) 1, 3
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
On Adversarial Vulnerability of Vision-Language Models through the Lens of Intermediate Spectral Subspaces
Aligning adversarial perturbations with the near-null singular directions of intermediate linear layers in transformer VLMs yields stronger attacks than existing feature- and output-space methods.