Low-Frequency Black-Box Backdoor Attack via Evolutionary Algorithm

arXiv:2402.15653v1 Announce Type: new
Abstract: While convolutional neural networks (CNNs) have achieved success in computer vision tasks, it is vulnerable to backdoor attacks. Such attacks could mislead the victim model to make attacker-chosen prediction with a specific trigger pattern. Until now, the trigger injection of existing attacks is mainly limited to spatial domain. Recent works take advantage of perceptual properties of planting specific patterns in the frequency domain, which only reflect indistinguishable pixel-wise perturbations in pixel domain. However, in the …

abstract algorithm arxiv attacks backdoor box cnns computer computer vision convolutional neural networks cs.cv domain low networks neural networks prediction spatial success tasks type via vision vulnerable