AI Data Science Summit 2022

Prof. Gal Chechik, Director of AI at NVIDIA, Professor at BIU

Title: Personalizing Federated learning with hypernetworks

Abstract: Federated learning (FL) is the task of learning a model over multiple disjoint local datasets. It is particularly useful when local data cannot be shared due to privacy, storage, or communication constraints. Personalized Federated Learning (PFL) allows each client to use a personalized model instead of a shared global model. The key challenge in PFL is to benefit from joint training while allowing each client to keep its own unique model and at the same time limit the communication cost. I’ll describe our recent work on using hypernetworks to personalize federated learning. This includes personalization to clients with new distributions, generalization to novel unsupervised clients at test time and generalization to clients with heterogeneous architectures.

Joint work with Ethan Fetaya, Or Litany, Aviv Navon, Aviv Shamsian, Idan Achituve, Haggai Maron, David Acuna, Sanja Fiedler, Jan Kautz

Tue, May 10, 2022

8:30 AM – 5:00 PM IDT