idanachituve / pfedgp Goto Github PK

Dear authors
Thank you for sharing source code, the flow of this paper is great and the code is also great.

Could you provide CINIC Dataset or source code about it?
I wonder Why is that dataset in paper and not in code. I wonder how you implement it.

Thank you for your time!

Dear author, firstly, thanks for sharing your awesome work.

If you don't mind me asking, in the case of pFedHN in your performance comparison table, the baseline that you compared to your amazing pFedGP, is this algorithm a version with a personalized classifier (pFedHN-pc)?

Dear author

In addition, there is a discussion that I am curious about your opinion.

In most cases, normalization is applied to datasets such as Cifar10 and Cifar100 in PFL.
In fact, I wonder if this normalization is justifiable in terms of not knowing the distribution of the entire dataset from the perspective of each client.

This discuss is a question I have about PFL overall methodologies.
This is a question I ask because I'm curious about your excellent point of view, apart from the github code.
Thank you always.

Firstly, thanks for your awesome work and code. When I run the pFedGP_IP_compute in the cifar100 in the 'noisy input', there raises an error: 'RuntimeError: cholesky_cuda: For batch 1: U(993,993) is zero, singular U.' The same problem also occurs in other settings. How can I solve the issue?