Deep Neural Networks: From Interpretation to Lightweight Design

发布者:569vip威尼斯游戏网站管理员发布时间:2021-11-25浏览次数:10

                Professor Maozhen Li

                Dept. of Electronic and Computer Engineering

                Brunel University London

                Uxbridge, UB8 3PH, UK

                Email: Maozhen.Li@brunel.ac.uk

                       

时间:2021年11月30日

腾讯会议:368280341


Talk Title:Deep Neural Networks: From Interpretation to Lightweight Design


Abstract

The past few years have seen a general development trend of deep neural networks (DNNs) in building deeper and larger scale models. These complex models require a large number of parameters and floating-point operations (FLOPs) while satisfying a certain level of accuracy in classification tasks, which is not conducive to deploying them on resource constrained mobile and embedded devices. In this talk, we review the research efforts on interpretation of DNNs with an aim to explain what a DNN has learned in the training process. We then introduce the concept of key features based on which we present CENet, a computationally efficient lightweight DNN that can be deployed potentially on mobile and embedded devices. CENet outperforms GhostNet, a CVPR’20 research work, from the aspects of both FLOPs in computation and accuracy in classification.


About the Speaker

Professor Maozhen Lireceived his PhD from the Institute of Software, Chinese Academy of Sciences in 1997. He did the Post-Doc research in the School of Computer Science and Informatics at Cardiff University, UK in 1999-2002. He is now a full Professor in the Department of Electronic and Computer Engineering at Brunel University London, UK. His research interests are in the areas of high-performance computing including cloud computing and edge computing, big data analytics, and intelligent systems with applications in smart grid and smart cities. He has over 180 scientific publications in these areas including 4 books and 90 peer reviewed journal papers. He has served over 30 IEEE conferences. He was the Chair of the TPC of FSKD’16, FSKD’14 and FSKD’12 respectively and he is on the Editorial Boards of a number of international journals. His collaborative research with Tongji University on Intelligent Transportation Systems was nominated by the Institution of Engineering and Technology (IET) for its Innovation Award in November 2015. He is a Fellow of the British Computer Society and the IET.



Baidu
sogou