[RFC] Found no advantages in model fusion methods

[xujli]

I use this framework to reproduce some algorithms which are helpful to accelerate convergence in iid and non-iid data distribution setting, such as FedAdp, Fedatt, FedProx, all these algorithms are in your ./examples dir. I ran these algorithms and plot the accuracy on test dataset, and found that the curves have almost no difference. There is no bug or execution confuse when I check the code. So I wonder if you can show me some examples and settings that these algorithms perform a more significant advance on convergence than FedAvg. It would be a great pleasure for me if you give me some brief explanations and advise to reproduce these algorithm.

Best Regards.

[baochunli]

This is related to the performance of these algorithms that are designed by the authors of their relevant papers, rather than issues of implementations in Plato. The implementations in examples/ are contributed by various contributors, and serve as references and examples for using Plato as a framework. One possible idea towards showing different results is to use more sophisticated models (rather than MNIST), and to introduce non-iid data distributions (using samplers provided by Plato) with the concentration parameter α defined appropriately in the corresponding Dirichlet distribution.

[xujli]

Thanks for your reply, I'll try more settings and models. I just want to get some advise on my results reproduction and I'm not familiar with the rules of Issues. I apologize for my wrong use. If I reproduce some other algorithms correctly, may I commit PR to the Repo?

[baochunli]

Of course.