Federated User Representation Learning: Privacy-Preserving Personalization in Distributed Systems

Explore Federated User Representation Learning (FURL) in this one-hour talk by Duc Bui, a PhD candidate at the University of Michigan. Delve into the challenges of personalization in Federated Learning (FL) and discover how FURL offers a simple, scalable, privacy-preserving, and resource-efficient solution for utilizing neural personalization techniques in FL settings. Examine the significant improvements in model quality achieved through FURL, with performance increases of 8% and 51% on two datasets, nearly matching centralized training results. Learn how FURL enables collaborative learning through shared parameters while maintaining user privacy, and gain insights into the structural similarities between user embeddings learned in FL and centralized settings.

Federated User Representation Learning