Prof. James T. Kwok, IEEE Fellow, Hong Kong University of Science and Technology, Hong Kong
Prof. Kwok is a Professor in the Department of
Computer Science and Engineering, Hong Kong University of Science and
Technology. He received his B.Sc. degree in Electrical and Electronic
Engineering from the University of Hong Kong and his Ph.D. degree in computer
science from the Hong Kong University of Science and Technology. Prof. Kwok
served/is serving as an Associate Editor for the IEEE Transactions on Neural
Networks and Learning Systems, Neurocomputing and the International Journal of
Data Science and Analytics. He has also served as Program Co-chair of a number
of international conferences, and as Area Chairs in conferences such as NIPS,
ICML, ECML, AAAI and IJCAI. He is an IEEE Fellow.
Speech: Compressing Deep Neural Networks
Abstract: Deep neural networks have been hugely successful in various domains, such as computer vision, speech recognition, and natural language processing. Though powerful, the large number of network weights leads to space, time and energy inefficiencies in both training and inference. In this talk, we will discuss recent attempts that try to reduce the model size. These include sparsification, quantization with fewer number of bits, low-rank approximation, distillation, and neural architecture search. They can greatly reduce the network size, and allow deployment of deep models in resource-constrained environments, such as embedded systems, smart phones and other portable devices.
Prof. Jianjun Li, Hangzhou Dianzi University, China
Prof. Jianjun Li received the PhD degree in Electrical and Computer Engineering from Windsor University, Canada. He is now serving as a chair professor of School of Computer Science and Technology in Hangzhou Dianzi University. He is also the director of Institute of Graphic and Image. Before this, Dr.Li worked in National Audiology Center (NCA) of Canada from 2003 to 2005, Mitsubishi Electronics Research Laboratory (MERL) of U.S.A from 2005 to 2006, École polytechnique fédérale de Lausanne (EPFL) of Switzerland from 2006 to 2007 as a visting scholar. He worked in Ambroda Ltd. for video coding stream processing of U.S.A from 2007 to 2010 as a senior engineer. From 2010 to 2012, Dr.Li worked as an assistant professor in Bilkent University and Ankara University, Turkey. In the meantime, he worked for FP-7 (now Horizon 2020) 3D project as a research fellow. Professor Li has worked in many different topics in computer vision, multimedia image processing, video coding and deep learning and published more than 50 papers in international conferences and journals and 2 books. He also has 3 contributions adopted by ISO/IEC Movie Picture Experts Group (MPEG) as a part of Reconfigurable Video Coding (RVC) standard. Dr.Li worked with International Institutes and Enterprises for more than 10 projects during his stay in abroad for more than 10 years. He now works on the National Science Foundation (NSFC) of China, National institutes and other Enterprises on more than 20 projects and holds 20 patents.
Prof. Li is also the recipient of several awards, including the “Qianjiang” scholar and the chief scientist of the innovation team of Zhejiang province in “3D industry and technology application”. Meanwhile, he is also a reviewer of many international journals and hold keynote speaker and committee member of many international conferences.
Speech: Gaze Estimation and Eye Track with CNN Networks
Abstract: Eye tracking and gaze estimation can improve the lives of people in many aspects, for example, people with motor disabilities by providing an intuitive and convenient input method, or provide detailed insights into users' attention. Also, gaze estimation in a non-intrusive manner can make human-computer interaction more natural, so it is also used in activity recognition, game and human-computer interaction, etc... Traditionally, eye tracking devices are expensive and have dedicated hardware, such as Tobii, Eyelink, SMI and so on. It has more needs for estimating gaze direction only from a low-resolution image of eyes. The difficulty of this task increases with background noise, less ideal illumination conditions, ambiguity, motion blur and a smaller number of pixels in determining the extent of the eyeball or shape of the iris. To determine the direction of gaze under the challenges: (1) low sensor quality or unknown/challenging environments, (2) large variations in eye region appearance, (3) lacking context or sense of distance/depth, and (4) physical differences between individuals, variations in head pose, or situational changes in decorations such as eyeglasses or cosmetics. It is appealing to adopt convolutional neural networks (CNN) with large amounts of data to solve the task of eye gaze direction estimation. Several datasets and works have been introduced in recent years.