In this paper, we explore methods to enhance both the performance and privacy of federated learning models by implementing two key techniques: homomorphic encryption and attention-based fusion.

Of course, if your PG server is running on a virtual machine, increasing hardware resources can also be an emergency option.