April 13, 2021 15:00 – 16:00
Tutorial I
Multi-Target and Multi-Camera Tracking under Deep Learning Detections (Prof. Jenq-Neng Hwang, Univ. of Washington, USA)
Abstract:
Thanks to the great advances on deep learning based multiple object detections, the multi-target tracking can follow the widely adopted tracking-by-detection framework, where the embedding features of the detected objects are used for frame-by-frame associations. Nonetheless, there are several critical issues need to be addressed under this framework, more specifically, the use of detections, the embedding features extraction, merging disjoint tracklets, use of instance segmentation masks, tracking across multiple cameras, camera link modeling, etc. In this one-hour tutorial, we will discuss the above issues by presenting some state-of-the-art approaches.
Biography
Dr. Jenq-Neng Hwang received the BS and MS degrees, both in electrical engineering from the National Taiwan University, Taipei, Taiwan, in 1981 and 1983 separately. He then received his Ph.D. degree from the University of Southern California. In the summer of 1989, Dr. Hwang joined the Department of Electrical and Computer Engineering (ECE) of the University of Washington in Seattle, where he has been promoted to Full Professor since 1999. He served as the Associate Chair for Research from 2003 to 2005, and from 2011-2015. He also served as the Associate Chair for Global Affairs from 2015-2020. He is the founder and co-director of the Information Processing Lab., which has won CVPR AI City Challenges awards in the past years. He has written more than 380 journal, conference papers and book chapters in the areas of machine learning, multimedia signal processing, and multimedia system integration and networking, including an authored textbook on “Multimedia Networking: from Theory to Practice,” published by Cambridge University Press. Dr. Hwang has close working relationship with the industry on multimedia signal processing and multimedia networking.
April 13, 2021 16:30 – 17:30
Tutorial II
How to Correctly Detect Face-Masks for COVID-19 from Visual Information? (Prof. Peter Peer, Univ. of Ljubljana, Slovenia)
Abstract
The new Coronavirus disease (COVID-19) has seriously affected the world. To limit the spread of the disease, mandatory face-mask rules are now becoming common in public settings around the world. Additionally, many public service providers require customers to wear face-masks in accordance with predefined rules (e.g., covering both mouth and nose) when using public services. These developments inspired research into automatic (computer-vision-based) techniques for face-mask detection that can help monitor public behaviour and contribute towards constraining the COVID-19 pandemic.
Although existing research in this area resulted in efficient techniques for face-mask detection, these usually operate under the assumption that modern face detectors provide perfect detection performance (even for masked faces) and that the main goal of the techniques is to detect the presence of face-masks only. In this tutorial, we revisit these common assumptions and explore the following research questions:
(i) How well do existing face detectors perform with masked-face images?
(ii) Is it possible to detect a proper (regulation-compliant) placement of facial masks?
iii) How useful are existing face-mask detection techniques for monitoring applications during the COVID-19 pandemic?
To answer these and related questions, we show comprehensive experimental evaluation of several recent face detectors for their performance with masked-face images. Furthermore, we show the usefulness of multiple off-the-shelf deep-learning models for recognising correct face-mask placement. Finally, we show a complete pipeline for recognising whether face-masks are worn correctly or not and compare the performance of the pipeline with standard face-mask detection models from the literature. To facilitate our study, we compiled a large dataset of facial images from the publicly available MAFA and Wider Face datasets and annotate it with compliant and non-compliant labels. The annotation dataset, called Face-Mask-Label Dataset (FMLD), is made publicly available to the research community.
The tutorial will, apart from answering the mentioned questions, put emphasise on scientific methodology for correct and meaningful comparison of approaches and repeatability of results, which includes having a large publicly available dataset with annotations, defining a training and testing protocol and performance measures. After all, majority of the attendees of the tutorial will be graduate students.
Biography
Peter Peer is a Full Professor at the Faculty of Computer and Information Science, University of Ljubljana, Slovenia, where he heads the Computer Vision Laboratory, coordinates the double degree study program with the Kyungpook National University, South Korea, and serves as a vice-dean for economic affairs. He received his PhD degree in computer science from the same institution in 2003. Within his post-doctorate he was an invited researcher at CEIT, Donostia – San Sebastian, Spain. He teaches courses on Operating systems, Game technology and virtual reality, and Image-based biometry. His research interests include biometrics and computer vision. He participated in several national and EU funded R&D projects and published more than 100 research papers in leading international peer reviewed journals and conferences. He is a co-organizer of the Unconstrained Ear Recognition Challenge (2017, 2019) and Sclera Segmentation Benchmarking Competition (2020). He serves as an Associated Editor of IEEE Access and IET Biometrics. He is a member of the EAB, IAPR and IEEE, where he also served as a chairman of the Slovenian IEEE Computer chapter for four years.
April 14, 2021 14:00 – 15:00
Tutorial III
Deep Learning for/in Wireless Networks: Fundamentals and Applications (Prof. Takayuki Nishio, Tokyo Institute of Technology, Japan)
Abstract
This tutorial aims to provide the fundamentals and applications of machine learning (ML), especially, deep supervised learning for wireless networks. The outline is as follows: (1) Brief introduction of basics of supervised learning, (2) Applications of deep supervised learning to the wireless networks, and (3) Federated learning: privacy-preserving deep learning in wireless networks. In detail, we address how to apply the machine learning techniques to challenges in wireless networks that are received power prediction and handover. Moreover, an emerging cooperative ML framework, namely federated learning, is introduced.
Biography
Takayuki Nishio received a Ph.D. degree in informatics from Kyoto University in 2013. He is currently an Associate Professor at the School of Engineering, Tokyo Institute of Technology. He provided tutorials on ML for wireless networks at IEEE ICC 2019, ITC32, and Japanese domestic conferences. His research interests are in wireless protocols, the applications of machine learning and computer vision to wireless networks, and communication-efficient machine learning mechanisms.
April 15, 2021 15:00 – 16:00
Tutorial IV
Multi-Agent Deep Reinforcement Learning for Connected and Autonomous Vehicles (Prof. Joongheon Kim, Korea Univ., Korea)
Abstract
In modern mobile and network research, there are active research progresses in connected and autonomous vehicle (CAV) algorithms design and implementation. Among various research contributions, the use of reinforcement learning algorithms is widely and actively considered in industry and academia societies.
The use of reinforcement learning algorithms is effective in CAV and mobile networks because it is optimal or pseudo-optimal in discrete-time stochastic sequential decision making under uncertainty, and thus, one of suitable solutions in time-varying communication network problem solving.
One of the most famous reinforcement learning algorithms, i.e., Markov decision process (MDP) guarantees optimal solutions using dynamic programming and the corresponding computational complexity is pseudo-polynomial, which means it provides optimal solutions when state spaces are relatively small. However, the state spaces are not small in modern applications, thus, deep neural networks are widely used for reinforcement learning approximation. This deep neural network based approximated reinforcement learning is called deep reinforcement learning (DRL).
However, it cannot represent cooperative behaviors among multiple agents. Thus, cooperative multi-agent DRL (MADRL) is essentially required, and thus, various MADRL algorithms have been proposed such as communication neural network (CommNet) and graph neural network with attention mechanism (G2ANet), CommNet is used because it is the MADRL algorithm for the case where all agents are fully associated and equivalent. G2ANet is also powerful if multiple heterogeneous agents have sophisticated relationship, whereas it introduces too much computation for our given homogeneous setting.
In this tutorial, various MADRL algorithms and applications for various CAV network problems such as charging, scheduling, and trajectory optimization. Lastly, this tutorial will lead active discussions for future research directions.
Biography
Joongheon Kim (M’06–SM’18) has been with Korea University, Seoul, Republic of Korea, since 2019, and he is currently an associate professor. He is also a vice director of Artificial Intelligence Engineering Research Center at Korea University, Seoul, Republic of Korea. He received the B.S. and M.S. degrees in Computer Science and Engineering from Korea University, Seoul, Republic of Korea, in 2004 and 2006, respectively; and the Ph.D. degree in Computer Science from the University of Southern California (USC), Los Angeles, California, USA, in 2014. Before joining Korea University, he was a research engineer at LG Electronics (Seoul, Republic of Korea, 2006–2009), member of technical staff at InterDigital (San Diego, CA, USA, 2012), systems engineer at Intel Corporation (Santa Clara in Silicon Valley, CA, USA, 2013–2016), and assistant professor at Chung-Ang University (Seoul, Republic of Korea, 2016–2019).
He is a senior member of the IEEE and serves as an associate editor for IEEE Transactions on Vehicular Technology. He internationally published more than 90 journals, 110 conference papers, and 8 book chapters. He also holds more than 50 patents, majorly for 60 GHz millimeter-wave IEEE 802.11ad and IEEE 802.11ay standardization.
He was a recipient of Annenberg Graduate Fellowship with his Ph.D. admission from USC (2009), Intel Corporation Next Generation and Standards (NGS) Division Recognition Award (2015), Haedong Young Scholar Award by KICS (2018), IEEE Vehicular Technology Society (VTS) Seoul Chapter Award (2019), Outstanding Contribution Award by KICS (2019), Gold Paper Award from IEEE Seoul Section Student Paper Contest (2019), Granite Tower Best Teaching Award by Korea University (2020), IEEE Systems Journal Best Paper Award (2020), and IEEE ICOIN Best Paper Award (2021).