Federated and Privacy-Preserving Learning

Machine-learning data are often distributed across users, organizations, or institutions and cannot readily be collected in a single location. We study federated and decentralized learning methods that enable collaborative model training under these constraints.

Our research addresses challenges arising from heterogeneous local data, limited labels, different local models, and the information exchanged during collaborative learning. We are particularly interested in understanding how useful models can be learned while reducing privacy risks associated with sharing data or model information.