Jul 2026
Spherical caps and feature cones: a geometric analysis of adversarial false-accepts
A closed-form geometric baseline for adversarial risk, an empirical measurement of deep face anisotropy using this baseline, and a demonstration that despite this “anisotropy shield,” gradient-based attacks remain feasible by exploiting the encoder’s local Jacobian are contributed.
Ilyes Azouani, Kévin Carta, Stéfane Mouille
· Symposium on Pattern Recogni... · 0 citations