Comparison of the Resilience of Convolutional and Cellular Neural Networks Against Adversarial Attacks

Convolutional architectures are commonly used in practice and the popularity of Cellular Neural Networks has also increased significantly in the past years, since these architectures can provide an efficient, low-energy, analogue implementation of deep neural networks.Both architectures were applied with success in various tasks, but in safety critical applications, like medical imaging or self-driving cars the high accuracy of these approaches is not enough, adversarial attacks still pose a significant threat in these areas.In this paper I will examine and compare how resilient these structures are towards various adversarial attacks and I will also investigate possible causes of these differences.