Keynote Talk 1: Bacteria Classification by Small-Scale Deep Learning

Abstract

Early classification of bacteria obtained from infected patients is of great importance for making a definitive diagnosis of patients and providing appropriate treatment. We have tried to classify bacteria using deep learning AI. We developed small-scale Depth-Wise Separable Convolutional Neural Networks (DCNNs). The layer structures of the DCNNs is much simpler than conventional Neural Networks (NN) structures. It has only 5 neuron layers, thereby reducing size of the NNs to 3.23 M parameters and 40.02 M MACs . The accuracy to classify bacteria was tested using DIBaS bacterial image dataset, and we have obtained accuracy of 96.28%