search

zshuai8 / FedGMM_ICML2023

Personalized Federated Learning under Mixture of Distributions
MIT License
15 stars 3 forks source link

n_clusters #3

Closed jxthyatt closed 9 months ago

jxthyatt commented 9 months ago

If we set n_clusters=3 (n_component=3) when generating non-iid clients, then theoretically how to set n_learner and n_gmm to get best results? is n_learner=3 and n_gmm=3 right? Please help to understand relations between the n_clusters and the n_learner, n_gmm?

zshuai8 commented 9 months ago

Depends on the input distribution of each cluster. If indeed n_cluster=3, and the input distribution is Gaussian, then n_gmm=3 is optimal. As for the relationship between the n_clusters and the n_learner, n_gmm, you might have to verify via empirical experiment results.