AutoML Models for Wireless Signals Classification and their effectiveness against Adversarial Attacks

In this paper, we study and compare the performance of AutoML models with state-of-the-art models on wireless signal classification and their vulnerability and transferability towards transfer-based white-box and black-box attacks. We designed models of four architectures using AutoML, namely Deep Residual Network (ResNet), Convolutional Long Short-Term Deep Neural Network (CLDNN), Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN). Using AutoML techniques for model generation helps to reduce time spent on designing, training and tuning hyper-parameters of deep learning models. Using numerical results, we show that AutoML models are a viable and solid candidate approach for the classification of wireless signals. In addition, we show the vulnerability of AutoML models towards adversarial attacks when compared to state-of-the-art models.