On The Resilience of Cellular Neural Networks to Low-intensity Adversarial Attacks

Abstract

Deep Neural networks are commonly used in various tasks and enabled the solution of many practical problems. These approaches usually result sufficiently high accuracy, but the robustness of these methods in critical applications is still under investigation. Adversarial attacks, in which minor perturbations can cause misclassification pose one of the most significant challenges. In case of convolutional neural networks there is ongoing research to create more resilient networks towards these attacks. In this paper I will demonstrate that multi-layered cellular neural networks in their nature are more robust and resilient to low-intensity attacks than their convolutional counterparts.