The Effective Depth Paradox: Topology and Trainability in Deep CNNs

Manfred M. Fischer, Joshua Pitts

This paper presents a controlled comparative study of convolutional neural network (CNN) topology and image classification performance across the architectural families VGG, ResNet…