Deep Network Classification by Scattering and Homotopy Dictionary Learning

We introduce a structured convolutional neural network which provides a simple model to analyze properties of deep representation learning and yields a higher classification accuracy than AlexNet over the ImageNet ILSVRC2012 dataset. This network is composed of a scattering transform which linearizes variabilities due to geometric transformations followed by a sparse l1 dictionary coding and a 2 hidden layer classifier. The whole pipeline is implemented in a deep convolutional network with a homotopy algorithm having an exponential convergence for the sparse l1 dictionary coding.