Squeeze and Excitation Block Based Neural Architecture Search with Randomization for CNN Construction

Neural Architecture Search (NAS) is a method for automatically designing neural networks. In this paper, we propose a block based NAS method that implements Randomization based transformation to generate the search space. The objective is to maximize the performance of a Convolutional Neural Network (CNN) model to detect the presence of tumors in mammography images, while minimizing the model size and complexity. The performance of two variants of our solution are compared. The first is based on Inception blocks and the second on a Squeeze and Excitation Inception block. The approaches are tested on the standard MNIST dataset as well as on the KAUBCMD mammogram dataset. 99.19% and 98.43% accuracy were achieved using the two datasets, respectively.