Sensitivity of Deep Convolutional Networks to Gabor Noise
This identifies a vulnerability in DCNs for security and robustness applications, but it is incremental as it builds on known UAP sensitivity.
The paper demonstrates that Deep Convolutional Networks (DCNs) are sensitive to Gabor noise patterns, which act as Universal Adversarial Perturbations (UAPs) across different architectures, revealing a poorly understood phenomenon.
Deep Convolutional Networks (DCNs) have been shown to be sensitive to Universal Adversarial Perturbations (UAPs): input-agnostic perturbations that fool a model on large portions of a dataset. These UAPs exhibit interesting visual patterns, but this phenomena is, as yet, poorly understood. Our work shows that visually similar procedural noise patterns also act as UAPs. In particular, we demonstrate that different DCN architectures are sensitive to Gabor noise patterns. This behaviour, its causes, and implications deserve further in-depth study.