Scalable, privacy-aware and (de)composable machine learning operations on the cloud-edge-far edge continuum
Research Seminar Series
-
Date
27 Jan 2025
-
Organiser
Department of Industrial and Systems Engineering, PolyU
-
Time
17:00 - 18:30
-
Venue
Online via ZOOM
Speaker
Dr Dimitris Chatzopoulos
Remarks
Meeting link will be sent to successful registrants
.jpg?as=1&sc=1&rev=89275a8c4e8b43899c95600de54cf8e3&hash=38E8821222752E1BDE849D7407080BDA)
Summary
The proliferation of AI enabled services deployed on various locations on the cloud-edge-far edge continuum has increased the management complexity of the resources utilized on the continuum due to their need for data collection and updates that adjust the models to environment changes (e.g., request type/distribution). In this talk, I will
- Use the application use-case of human activity recognition to demonstrate how decomposable ML operations can achieve higher accuracy when deployed in the whole spectrum of the continuum.
- Introduce a parallel split learning-based protocol to demonstrate how the required resources for (re)training multiparameter models can be reduced substantially.
- Introduce a microservice approach based on reinforcement learning to show how resources on the continuum can be managed more effectively.
Keynote Speaker
Dr Dimitris Chatzopoulos
Assistant Professor and Ad Astra fellow
School of Computer Science, University College Dublin and Insight SFI Research Centre for Data Analytics, Ireland
Dimitris Chatzopoulos is an Assistant Professor and Ad Astra Fellow at the School of Computer Science of University College Dublin (UCD) and a primary investigator at the Insight Science Foundation Ireland Research Centre for Data Analytics. He is the director of the decentralized AI systems group that is currently composed of eight members and has received significant financial support (around 10M HKD) from the European Union (MLSysOps and DIGITISE Horizon Europe projects) and the industry (Qualcomm). Before moving to UCD in October 2021, he was a Research Assistant Professor in the Department of Computer Science and Engineering of the Hong Kong University of Science and Technology. Dimitris received his PhD in Computer Science and Engineering from The Hong Kong University of Science and Technology and his Diploma and MSc in Computer Engineering and Communications from the University of Thessaly, Greece. In 2014 he was a visiting researcher in EPFL and in 2016 in Tsinghua University and in Cambridge University. In 2017 he was selected to participate in the 5th Heidelberg Laureate Forum. His research interests include privacy-preserving and AI-enabled decentralized applications for mobile and distributed systems. Dimitris has published his research in top-tier conferences and journals such as (IEEE Transactions on Mobile Computing, IEEE Infocom, ACM IMWUT, IEEE Transactions on Dependable and Secure Computing, and others)
You may also like
