FedProx, only the performance about clients' own testset was concerned, without global testset. I know that sometimes personalized FL concerns about client's testset, but why not do we compare personialized FL with local train in clients?
If we just concern about clients' own testset, I think comparsion experiments with the profermance of local train are necessary.
FedProx, only the performance about clients' own testset was concerned, without global testset. I know that sometimes personalized FL concerns about client's testset, but why not do we compare personialized FL with local train in clients? If we just concern about clients' own testset, I think comparsion experiments with the profermance of local train are necessary.