Optimizing Client Interactions in Collaborative Learning through Combinatorial Approaches

Abstract

In collaborative learning with highly heterogeneous data distributions, optimizing client interactions is crucial for enhancing both personalization and generalization among participants. This talk explores the application of combinatorial optimization in collaborative learning to effectively group or prioritize clients, thereby improving their capabilities. By framing client interactions as a combinatorial problem, we can derive solutions that lead to significant performance enhancements. Case studies and empirical results illustrate how these methodologies elevate learning outcomes across diverse collaborative environments.

Bio

Salma Kharrat is a Ph.D. candidate in Computer Science at KAUST, specializing in federated learning and optimization. In 2023, she was recognized with the Dean’s List Award for her outstanding academic performance. During her doctoral studies, Salma had the opportunity to intern at EPFL. She holds an M.Sc. in Computer Science (2022) and a Diplôme d’Ingénieur in Computer Science and Engineering from the Tunisian National School of Computer Science (2020)