Exploiting Time and Space in Federated Learning
Malcolm Egan, INRIA
3-4pm 9th Jul 2025
Abstract
Environmental monitoring applications often involve a large number of sensing devices, each continuously observing a physical process (e.g., climate variables). On the other hand, each device can only observe its local environment which does not provide data representative of a larger region. Communication of large quantities of data from sensors to a centralized server is well known to be a challenging problem. Federated learning can therefore play a key role to construct good models while significantly reducing communication requirements. In this talk, I will present recent work (https://arxiv.org/pdf/2503.18807) characterizing the convergence of federated learning algorithms with streaming data. As physical processes are often non-i.i.d. in time and nearby sensors will typically observe correlated samples, it is important to understand how temporal and spatial dependence impacts the quality of the learned model. These features of the data have important implications for the choice of algorithm and, as we will also briefly discuss, communication strategies.
Short Bio
Malcolm Egan received the Ph.D. in Electrical Engineering in 2014 from the University of Sydney, Australia. He is currently a Chargé de Recherche (Tenured Research Scientist) in Inria and a member of CITI, a joint laboratory between Inria, INSA Lyon and Université de Lyon, France. Previously he was an Assistant Professor in INSA-Lyon, and a postdoctoral researcher with the Laboratoire de Mathématiques, Université Blaise Pascal, France and the Department of Computer Science, Czech Technical University in Prague, Czech Republic. He was an associate editor for IEEE Communications Letters and is currently an editorial board member for Nature Scientific Reports. He is also a recipient of the French National Research Agency (ANR) Young Researcher (JCJC) Grant. His research interests are in the areas of information theory, statistical signal processing and machine learning with applications in wireless communications and modeling of biochemical systems.