Why Deep Learning Networks Work So Well?
Professor C.-C. Jay Kuo, IEEE Fellow, AAAS Fellow, SPIE Fellow
University of Southern California, USA

Deep learning networks, including convolution and recurrent neural networks (CNN and RNN), provide a powerful tool for image, video and speech processing and understanding nowadays. However, their superior performance has not been well understood. In this talk, I will unveil the myth of CNNs. To begin with, I will describe network architectural evolution in three generations: first, the McClulloch and Pitts (M-P) neuron model and simple networks (1940-1980); second, the artificial neural network (ANN) (1980-2000); and, third, the modern CNN (2000-Present). The differences between these three generations will be clearly explained. Next, theoretical foundations of CNNs have been studied from the approximation, the optimization and the signal representation viewpoints, and I will present main results from the signal processing viewpoint. A good theoretical understanding of deep learning networks provides valuable insights into the past, the present and the future of their research and applications.

Speaker’s Biography
Dr. C.-C. Jay Kuo received his Ph.D. degree from the Massachusetts Institute of Technology in 1987. He is now with the University of Southern California (USC) as Director of the Media Communications Laboratory and Dean’s Professor in Electrical Engineering-Systems. His research interests are in the areas of digital media processing, compression, communication and networking technologies. Dr. Kuo was the Editor-in-Chief for the IEEE Trans. on Information Forensics and Security in 2012-2014. He was the Editor-in-Chief for the Journal of Visual Communication and Image Representation in 1997-2011, and served as Editor for 10 other international journals. Dr. Kuo received the 1992 National Science Foundation Young Investigator (NYI) Award, the 1993 National Science Foundation Presidential Faculty Fellow (PFF) Award, the 2010 Electronic Imaging Scientist of the Year Award, the 2010-11 Fulbright-Nokia Distinguished Chair in Information and Communications Technologies, the 2011 Pan Wen-Yuan Outstanding Research Award, the 2014 USC Northrop Grumman Excellence in Teaching Award, the 2016 USC Associates Award for Excellence in Teaching, the 2016 IEEE Computer Society Taylor L. Booth Education Award, the 2016 IEEE Circuits and Systems Society John Choma Education Award, the 2016 IS&T Raymond C. Bowman Award, and the 2017 IEEE Leon K. Kirchmayer Graduate Teaching Award. Dr. Kuo is a Fellow of AAAS, IEEE and SPIE. He has guided 140 students to their Ph.D. degrees and supervised 25 postdoctoral research fellows. Dr. Kuo is a co-author of about 250 journal papers, 900 conference papers and 14 books.
The Active Efficient Coding Framework for the Joint Emergence of Perception and Behavior
Professor Bertram E. Shi, IEEE Fellow
Head of Department of Electronic and Computer Engineering
Hong Kong University of Science and Technology, Hong Kong

Based on sensory information, biological and robotic agents create perceptual representations encoding the environmental state, and use these representations as the basis for generating actions in the environment. In robotic systems, both the representations and their mapping to action are typically hard coded, built upon explicit knowledge about the geometry of the sensors and actuators, often obtained through precise calibration. Although learning can be used, past efforts have been focused on learning either the perceptual representation (e.g. via supervised learning) or action generation (e.g. via reinforcement learning) in isolation and often restricted to an initial training stage. In contrast, biological systems demonstrate remarkable lifelong adaptability in both perception and action, even in the presence of significant changes in body shape and function, e.g. due to growth or injury. We suggest that the simultaneous plasticity of both the perceptual representation and its mapping to action is one of the key reasons for the robustness of biological systems in comparison with their robotic counterparts. Modelling this process will not only improve our understanding, but also enable robotic systems to exhibit similar degrees of lifelong adaptability. As a step towards this goal, we describe the "active efficient coding" (AEC) hypothesis, an extension of Barlow’s efficient coding hypothesis. AEC posits that organisms adapt both their sensory processing and their behavior to best represent the environment under limited resource constraints. As a concrete examples of this framework, we will describe its application to the unsupervised learning of a wide variety of eye/camera movements observed in/utilized by humans/robots. These include smooth pursuit/tracking, binocular vergence, and saccadic eye movements driven by attentional mechanisms. Interestingly, the AEC provides a single framework that accounts for all of these. The model is sufficiently details that we can use it to replicate psychophysical experiments on humans in the literature obtaining similar results. It can also be used directly in robotic platforms for robust and automatic multisensory cue integration.

Speaker’s Biography
Dr. Bertram E. Shi received his B.S. and M.S. degrees in Electrical Engineering from Stanford University in 1987 and 1988, and the Ph.D. degree in Electrical Engineering from the University of California at Berkeley in 1994. He then joined the faculty at the Department of Electronic and Computer Engineering at the Hong Kong University of Science and Technology, where he currently serves as Department Head and Professor with joint appointment in the Division of Biomedical Engineering. His research interests are in bio-inspired signal processing and robotics, neuromorphic engineering, computational neuroscience, developmental robotics, machine vision, image processing, and machine learning. Prof. Shi is an IEEE Fellow and served as Distinguished Lecturer for the IEEE Circuits and Systems Society twice. He has also served as on the editorial boards of the IEEE Transactions on Circuits and Systems, the IEEE Transactions on Biomedical Circuits and Systems and Frontiers in Neuromorphic Engineering. He served as Chair of the IEEE Circuits and Systems Society Technical Committee on Cellular Neural Networks and Array Computing and as General Chair and Technical Program Chair of conferences in that area.
When Optical Spectra Meet Big Data for Wireless Communications
Professor Zhengyuan Xu, Thousand Talents Program of China
University of Science and Technology of China, China

Wireless transmission of wide optical spectrum signals explores the rich optical spectra from infrared to visible light and ultraviolet bands for reliable and high rate wireless communications. Its unique features of unlicensed spectrum and anti-electromagnetic interference help to meet the needs of the future enhanced mobile broadband services, massive connections of things, and ultra-low latency communications. However, those new services and applications may concurrently generate and support Wireless Big Data from heterogeneous wireless terminals and networks. This talk will cover communication and signal processing aspects of optical wireless communications, and demonstrate how wireless communication system design can be driven by big data when both wide optical spectra and a huge wireless data set are available.

Speaker’s Biography
Professor Zhengyuan Xu received his B.S. and M.S. degrees from Tsinghua University, Beijing, China, in 1989 and 1991, respectively, and Ph.D. degree from Stevens Institute of Technology, New Jersey, USA, in 1999. From 1991 to 1996, he was with Tsinghua Unisplendour Group Corporation, Tsinghua University, as system engineer and department manager. In 1999, he joined University of California, Riverside, first as Assistant Professor and then tenured Associate Professor and Professor. He was Founding Director of the multi-campus Center for Ubiquitous Communication by Light (UC-Light), University of California. In 2010, he was selected by the “Thousand Talents Program” of China, appointed as Professor at Tsinghua University, and then joined University of Science and Technology of China (USTC) in 2013. He is Founding Director of the Optical Wireless Communication and Network Center, Founding Director of Wireless-Optical Communications Key Laboratory of Chinese Academy of Sciences, in USTC. He is also a chief scientist of the National Key Basic Research Program (973 Program) of China, and principal investigator of the clustered key projects on Wireless Big Data from National Science Foundation of China. His research focuses on wireless communications and networking, optical wireless communications, geolocation, intelligent transportation, and signal processing. He has published over 270 journal and conference papers, and co-authored a book titled Visible Light Communications: Modulation and Signal Processing, to be published by Wiley-IEEE Press. He has served as an Associate Editor and Guest Editor for different IEEE and OSA journals. He was a Founding Chair of IEEE Workshop on Optical Wireless Communications in 2010.
Gated Deep Neural Networks for Adaptive Information Flow
Dr. Gang Wang
Chief Scientist
Alibaba AI Lab, China

Human brains are adept at dealing with the deluge of information they continuously receive, and adaptively controlling and regulating the information flow to focus on the important inputs and suppress the non-essential ones for better performance. Inspired by such a capability, we develop three types of networks which computationally regulate the information flow in CNN, siamese CNN, and LSTM respectively. Our methods have achieved state-of-the-art performance on CIFAR 100 for image classification, Market-1501 dataset for human re-identification, and NTU RGB-D dataset for action recognition.

Speaker’s Biography
Wang Gang is currently a researcher/senior director and a distinguished scientist in Alibaba AI Labs. He was an Associate Professor with the School of Electrical and Electronic Engineering at Nanyang Technological University (NTU). He had a joint appointment at the Advanced Digital Science Center (Singapore) as a research scientist from 2010 to 2014. He received his B.Eng. degree from Harbin Institute of Technology in Electrical Engineering and the PhD degree in Electrical and Computer Engineering, University of Illinois at Urbana-Champaign. He is a recipient of MIT technology review innovator under 35 award (Asia). He is an associate editor of TPAMI and an area chair of ICCV 2017 and CVPR 2018.

Contact us:
Copyright © ISPACS 2017 All rights reserved.