Title: Universality of deep CNNs and distributed learning
Authors: Ding-Xuan Zhou - City University of Hong Kong (Hong Kong) [presenting]
Abstract: The aim is to show that deep convolutional neural networks (CNNs) with the rectified linear unit activation function without any fully-connected layers are universal as the level tends to infinity. We shall also discuss some problems related to distributed learning. Our approach is based on machine learning and approximation theory.