Information-Geometric Inverse Distillation for Enhancing Adversarial Transferability
It is proved that, under a matched fixed-anchor implementation, soft-label cross-entropy and KL divergence differ only by a constant entropy term and therefore induce identical gradients, Hessians, and adversarial optimization trajectories.
Wenyuan Wu, Yuan Sun, Ying-Ke Chen et al.
· 2 citations