Date and Time: Wednesday, 13th August 2025, 11.00AM
Venue: New Seminar Room

Title of the Talk: 
Towards Sustainable AI: Data-driven Energy Management in Digital Infrastructures

Abstract of the Talk: 
Abstract: 
Contemporary large-scale computing systems are becoming increasingly heterogeneous and 
decentralized, driven by the growing demand for digitization, particularly due to the 
recent surge in low-latency AI/GenAI applications. This trend is exemplified by the 
deployment of edge micro Data Centers (DCs) alongside existing hyper-scale cloud DCs. 
These cyber-physical infrastructures consume substantial energy and significantly contribute 
to carbon emissions. Assessing and managing energy use in such large, distributed systems 
presents numerous challenges, including hyper-heterogeneity of hardware and power 
characteristics, diversity of user workloads and software stacks, and potential 
performance degradation from energy-saving measures.
In this talk, I will present our recent research on data-driven modeling and optimization 
techniques for energy-aware management of large-scale cloud and edge systems. I will 
illustrate how these methods address heterogeneity, balance performance with sustainability 
objectives, and significantly reduce the energy footprint of AI workloads, 
moving closer to the goal of sustainable AI.