Federated Learning of Generative Adversarial Networks with Resource Constraints and Unreliable Communication
This project develops federated learning schemes for generative AI that address the challenges of centralized training in dynamic, sensitive environments. By co-designing algorithms and systems, it enables adaptive, asynchronous learning and incentivizes participation across heterogeneous, unreliable clients. The approach is designed to ensure robust model training while accounting for real-world constraints in communication, computation, and data distribution.