I am currently working on a personalized federated learning research. The current simulation structure is as follows:
Global model with with 5 global rounds
Local model with 10 clients and 4 local rounds
Model is CNN for 11 classes classification.
While the code is running, the local models have accuracies all over the place. Here is an example:
Is this normal for personalized federated learning? I can also observe this behavior in all datasets that I have tried so far.
I do want to share more details with you, but since this is an ongoing research, I am not sure how much details I can share.
I have not tried anything else yet since my knowledge about federated learning is very limited, and I am not sure if there is an actual issue in first place.
