Title: Distributed Machine Learning and Network Resource Allocation for Intelligent Edge Services

Speaker: Prof. Iordanis Koutsopoulos (Athens University of Economics and Business)

Abstract: In this talk, we will outline some research areas we work on in the last few years with the aim to build the foundations of intelligent and trustworthy services at the resource-limited network edge. These problems require new synergies between Artificial Intelligence (AI) and Networking, and they constitute pieces of the puzzle of ensuring trustworthy services through properties such as low-latency inference, personalization, explainability, fairness, continual model updates, energy-efficiency, and low-latency communication with the network controller. The areas include: (1) running Large Language Models (LLMs) over a network of resource-constrained devices for low-latency inference; (2) developing new architectures for Split Federated Learning, by splitting and parallelizing the workload of the training phase across devices; (3) achieving exact optimization of the training objective in personalized Federated Learning; (4) addressing explainability and accuracy during training through a Multi-objective optimization framework; (5) introducing class fairness in Continual Learning; (6) using Reinforcement Learning to rethink the problem of issuing energy usage recommendations; (7) leveraging Multi-armed bandits to learn the optimal controller placement and optimal routing policies in Software-Defined Networks, and the optimal offloading policy for image processing tasks.

We will also outline some recently completed and active European projects in different verticals such as health, autonomous vehicles, mobile ad-hoc networking, energy, agriculture that oftentimes fuel and provide the use cases for the problems above. We will conclude with some problems we have recently started working on in quantum networking and will pave the way ahead.

Short Biography: Iordanis Koutsopoulos is Professor at the Department of Informatics of Athens University of Economics and Business (AUEB) since 2021. He obtained the Diploma in Electrical and Computer Engineering from the National Technical University of Athens (NTUA), Athens, Greece, in 1997, and the M.Sc. and Ph.D. degrees in Electrical and Computer Engineering from University of Maryland College Park (UMCP) in 1999 and 2002 respectively. He has been with AUEB since 2013; prior to that he has been with the Department of Electrical & Computer Engineering of University of Thessaly, Volos, Greece (2005-2012).

During his sabbatical in 2012 he was a visiting research scientist with Yahoo! Research Labs, Barcelona, Spain. During 2005, he held a visiting scientist position with the University of Washington, Seattle, USA. He has also held internship and research scientist positions in USA, at Hughes Network Systems, MD, Hughes Research Labs, CA, and Aperto Networks, CA.  In 2015, he was awarded a European Research Council (ERC) runner-up award for a project on Resource Management for Self-coordinated Autonomic Wireless Networks.

Together with his collaborators, he has co-authored about 200 research papers which have attracted 6.320 citations, forming an h-index of 36. He has received 3 best-paper awards for papers in the IEEE ECOFEC workshop 2019 (in conjunction with IEEE INFOCOM), in the ACM HotPost workshop 2016 (in conjunction with ACM Mobihoc), and in the ACM WinTECH 2013 workshop (in conjunction with ACM Mobicom). Since 2020, he has been included in each year’s list of World’s top 2% scientists, published by Stanford University.

His research interests are in the intersection of applied AI and networked systems control and optimization, with applications to wireless networks, edge computing, and different verticals. He has attracted about €7.2M of funding from European, national and private funding sources and has been PI and technical coordinator for several European and national projects. He currently serves as the Director of the M.Sc. program on Computer Science in the Department of Informatics of AUEB.