-
Notifications
You must be signed in to change notification settings - Fork 155
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Are the procedures of weights aggregation and clients selection correct? #10
Comments
Thanks for your questions! There are at least two sampling schemes we can consider in this problem: (1) sampling devices with weights proportional to the number of local data points and simply averaging the updates, and (2) sampling devices uniformly and doing weighted aggregation with weights proportional to the number of local data points. In order to perform a fair comparison with FedAvg, we use the second scheme, which is slightly different from our algorithm (see Section 5.1 -- Implementation of the paper). Interestingly, the proposed algorithm (using the first sampling scheme) performs slightly better for both FedProx and FedAvg (see Figure 12 in Appendix C.3.4). That's a good point. Please let me know if you have other questions! |
Thank you for your reply. I have implemented the schemes following your instruction. I have another question: Did you consider the system heterogeneity when compare the differences between scheme 1 and 2? |
For Figure 12 in Appendix C.3.4, no. But you can always do that, just that we were concerned about the statistical efficiency of different sampling schemes, so we didn't mix it with the systems heterogeneity. And Fedprox is agnostic of the sampling schemes and can yield similar improvement under either scenario. |
Thank you! I will try it. |
Good job. I have read the paper and codes. I have some questions:
Thank you.
The text was updated successfully, but these errors were encountered: