Prof. Huaglory Tianfield
Professor of Computing, PhD
Director, AI + IoT Research Lab
Glasgow Caledonian University
Scotland, United Kingdom
Title: Decentralised Edge Intelligence with Federated Learning and Blockchain Technology
Bio:
Huaglory Tianfield has been a Professor of Computing with Glasgow Caledonian University, Scotland, United Kingdom since March 2001. Prof Tianfield is extensively involved in professional activities. He is member of EPSRC Peer College, Chair of Technical Committee on Cyber-Physical Cloud Systems of IEEE Systems, Man, and Cybernetics Society, Editor-in-Chief of Multiagent and Grid Systems – An International Journal of Data Science and Engineering. Prof Tianfield directs the AI + IoT Research Lab. His research areas include Trustworthy AI, Human-Centred AI, Federated Learning, Cloud Computing, and AI for Precision Health. He is (co-)author to over 200 research articles published in refereed journals and conferences, and is a frequent invited speaker at conferences and institutions all over the world. Hua Tianfield earned his Bachelor of Engineering (Hons), Master of Engineering and Doctorate of Engineering degrees all in Electronic Engineering (Industrial Automation).
Abstract:
With the confluences in edge computing and Internet of Things (IoT) big data, decentralised edge intelligence has emerged as an innovative paradigm that combines trusted collaboration, Distributed Artificial Intelligence (DAI), and decentralised deployment of AI. This talk will provide an in-depth exploration on how decentralised edge intelligence is empowered through federated machine learning and blockchain technology.
We will first elucidate the concept of decentralised edge intelligence and highlight its significance in the era of scalable, efficient, secure collaboration over edge computing and IoT big data. Then, we will look into the systems architectures of decentralised edge intelligence, discussing the benefits of distributed processing, such as reduced latency, improved scalability, and enhanced privacy.
Federated learning is a best fit to decentralised edge intelligence. We will investigate how federated learning enables collaborative model training across multiple edge devices, thus preserving data privacy. The vulnerabilities associated with federated learning, including data security and model poisoning attacks, will also be addressed. Furthermore, we will explore the integration of blockchain technology and smart contracts into decentralised edge intelligence. These technologies consolidate the robust and secure frameworks in decentralised edge intelligence systems.
Whilst extremely exciting, decentralised edge intelligence is still a very dynamic area. Prior to conclusion, we will seriously look into the main open issues and challenges faced by this emerging paradigm of decentralised edge intelligence.
Prof. Chong-Yung Chi
National Tsing Hua University, Hsinchu, Taiwan
IEEE Fellow and an AAIA Fellow
Title: Convergence of Convex Optimization & Artificial Intelligence
Bio:
Chong-Yung Chi (祁忠勇) received B.S. degree from Tatung Institute of Technology, Taipei, Taiwan in 1975, Master degree from National Taiwan University, Taipei, Taiwan in 1977, and Ph.D. degree from the University of Southern California, Los Angeles, California, in 1983 all in Electrical Engineering. Currently, he is Professor of National Tsing Hua University, Hsinchu, Taiwan. He has published more than 240 technical papers (with citations more than 7,000 times by Google-Scholar), including more than 90 journal papers (mostly in IEEE Trans. Signal Processing), more than 140 peer-reviewed conference papers, and 2 books, including a textbook, Convex Optimization for Signal Processing and Communications from Fundamentals to Applications, CRC Press, 2017 (which has been popularly used in a series of invited intensive short courses at 10 top-ranking universities in Mainland China since 2010 before its publication). He received 2018 IEEE Signal Processing Society Best Paper Award, entitled “Outage Constrained Robust Transmit Optimization for Multiuser MISO Downlinks: Tractable Approximations by Conic Optimization,” IEEE Trans. Signal Processing, vol. 62, no. 21, Nov. 2014. His current research interests include signal processing for wireless communications and networking, convex analysis and optimization for blind source separation, biomedical and hyperspectral image analysis, graph based learning and signal processing, and data security and privacy protection in machine learning.
He is an IEEE Fellow and an AAIA Fellow. He has been a Technical Program Committee member for many IEEE sponsored and co-sponsored workshops, symposiums and conferences on signal processing and wireless communications, including Co-Organizer and General Co-Chairman of 2001 IEEE Workshop on Signal Processing Advances in Wireless Communications (SPAWC). He was an Associate Editor (AE) for four IEEE Journals, including IEEE Trans. Signal Processing for 9 years (5/2001~4/2006, 1/2012~12/2015), and he was a member of Signal Processing Theory and Methods Technical Committee (SPTM-TC) (2005-2010), a member of Signal Processing for Communications and Networking Technical Committee (SPCOM-TC) (2011-2016), and a member of Sensor Array and Multichannel Technical Committee (SAM-TC) (2013-2018), IEEE Signal Processing Society.
Abstract:
Mathematical optimization, such as convex optimization (CVX-Opt), that has been extensively applied in sciences and engineering over the last decades. Artificial Intelligence (AI), such as Machine Learning (ML) and Deep Learning, has been pervasive not only in sciences and engineering, but also in our daily life. Both CVX-Opt and AI currently perform outstandingly and independently in many different fields with surprising application advantages, in spite of some existing challenges yet to be resolved. For the former, with no need of training data, a specific mathematical model and problem formulation is required, and practical or acceptable approximate solutions can always be practically obtained, together with insightful performance analysis that may be achievable mostly. For the latter, big training dataset and tremendous computing complexity are frequently required, though no intricate mathematics. In this speech, we will address their convergence (i.e. their wonderful fusion or combination), potentially with fantastic benefits in learning performance, running time saving, and problem scalability, etc. We will present the convergence of CVX-Opt and ML and/or DL by the following instances:
- Computing Resource Allocation in Space-Aerial Integrated Network
- Federated Learning for Data Classification and Clustering with Privacy Protection
- Hyperspectral Image Denoising and Inpainting
Finally, we draw some conclusions as well as some future research explorations.