Custom loss function are implemented in order to realize the federated algorithms.
The FedProx Custom Loss adds a proximal term to the local function to restrict local updates to be closer to the initial global model.
The FedDyn Custom Loss optimizes a local empirical risk objective, which is the sum of its local empirical loss and a penalized risk function. The penalized risk, which is dynamically updated, is based on current local device model, and the received server model.
