arrow drop search cross

Seminar - "Combining Complex Wavelets with Deep Networks: aiming to improve learning efficiency for vision systems"

by Prof. Nick Kingsbury, University of Cambridge, Dept. of Engineering, UK


14/01/2019   :   14h00
Inria Sophia Antipolis Méditerranée
Publication : 14/01/2019
Partager cet article :

Scattering networks [Bruna & Mallat, IEEE Trans PAMI 2013; Oyallon & Mallat, CVPR 2015] may be interpreted as convolutional network layers in which the filters are defined by complex wavelet transforms and whose layer non-linearities are typically complex modulus (L2-norm) operators.

Usually they are pre-designed using standard complex wavelet design methodologies that are based on accumulated human knowledge about vision systems, and they involve minimal training. It is found that several layers of scatternet can usefully replace the early layers of a deep convolution neural net (CNN). The aim of this strategy is that the deterministic and complete nature of the wavelet transformations will result in deep networks that are faster at learning, more comprehensible in their behaviour and perhaps better at generalisation than a CNN which has to learn all of its layers from finite amounts of training data.

Furthermore, by employing tight-frame overcomplete wavelets and L2-norm nonlinearities, signal energy may be conserved through the scatternet layers, leading to some interesting strategies for subspace selection.

In this talk we shall suggest a number of ways that dual-tree complex wavelets may be incorporated into deep networks, either to generate scatternet front-ends or to produce interesting alternatives to standard convolutional layers, embedded deeper in the network. We will also show how recent ideas on CNN layer visualisation can be extended to include the wavelet-based layers too. We shall pose more questions than answers, while also presenting a few results from current stages of this work. I am very grateful to my co-researchers on this project, Amarjot Singh and Fergal Cotter.

 

Speaker:

Nick Kingsbury received the honours degree in 1970 and the Ph.D. degree in 1974, both in electrical engineering, from the University of Cambridge. He is a Fellow of the IEEE. From 1973 to 1983 he was a Design Engineer and subsequently a Group Leader with Marconi Space and Defence Systems, Portsmouth, England, specializing in digital signal processing and coding theory. Since 1983 he has been a Lecturer in Communications Systems and Image Processing at the University of Cambridge and a Fellow of Trinity College, Cambridge. He was appointed to Professor of Signal Processing in 2007 and was head of the Signal Processing and Communications Research Group until 2016. In 2017 he retired from departmental teaching and admin responsibilities in order to pursue research as a Fellow of Trinity College.

His current research interests include image analysis and enhancement techniques, object recognition, motion analysis and registration methods. He has developed the dual-tree complex wavelet transform and is especially interested in the application of wavelet frames to the analysis of images and 3-D datasets, including their use in deep learning networks.


Location: EULER Violet room, Inria Sophia-Antipolis Méditerranée