Exposé de Salim El Rouayheb

mardi 2 juillet 2024
On Tuesday, July 2nd at 2:00 pm, Prof. Salim El Rouayheb, associate professor at Rutgers University, New Jersey, USA will be giving a talk on decentralized learning in conference room 007 at the i3S lab.
The talk entitled « Walk for Learning : A Random Walk Approach for Decentralized Learning on Graphs »
Abstract
Distributed learning methods like Federated Learning rely heavily on a central server, known as the Parameter Server (PS), to aggregate and disseminate model updates in each iteration. This dependence on the PS creates a single point of failure, making the system vulnerable when the PS fails, is unavailable, or poses security and privacy risks. Additionally, communicating with the cloud-based PS can become a significant bottleneck, especially when dealing with large models. As an alternative, we explore Random Walk (RW) learning, which enables distributed learning without relying on a central server. The setting consists of data that is distributed across the nodes of a graph, and the goal is to learn a global model of the distributed data without the standard reliance on a central server. We investigate a decentralized SGD algorithm in which a random walk on the graph carries a global model that is updated based on the local data of the visited node. We focus on designing RW strategies to accelerate algorithm convergence using importance sampling techniques.
Biography
Salim El Rouayheb is an associate professor in the ECE Department at Rutgers University. From 2013 to 2017, he was an assistant professor at the ECE Department at the Illinois Institute of Technology, Chicago. He was a research scholar at the Electrical Engineering Department at Princeton University (2012-2013) and a postdoc at the EECS department at the University of California, Berkeley (2010-2011). He received his Ph.D. degree in Electrical Engineering from Texas A&M University, College Station, in 2009. In 2019, he was the Rutgers University Walter Tyson Junior Faculty Chair. He received the Google Faculty Award in 2018 and the NSF CAREER award in 2016. His research interests lie in the area of information-theoretic security and privacy of data in networks and distributed systems.