Skip to content
InfoResearchPreprint

Density Feedback Control for Transferability of Generative Sparse Adversarial Attacks

Published
Record updated
View JSON

Summary

Density Feedback Control (DFC) is proposed to fix the perturbation density of generator-based sparse adversarial attacks, which otherwise depends on a fixed sparsity coefficient whose density mapping is unknown and shifts during training. DFC adjusts the coefficient at the end of each epoch based on the gap between observed hard-mask density and the target, without changing the generator architecture or loss. At a 10% target density, it brings three base attacks within one percentage point of target in a single run and improves transfer attack success rate for two of them on CNN and Transformer targets, while perturbing fewer pixels. It yields no consistent gain for EGS-TSSA, which restricts perturbations to source-model feature-importance regions.