Review for NeurIPS paper: An Efficient Framework for Clustered Federated Learning

Additional Feedback: Empirical Analysis: - The approach is not compared to related work. Straight-forward baselines would be clustering on the central machine approach [9] or the fine-tuning of global models [7, 35] which are cited in the paper. Theoretical Analysis: My main concern with the theoretical analysis is the assumption that initial models are already very close their correct clusters (1/4 of the minimum distance between cluster centers for the linear models - for the strong convex problems an additional factor comes in that depends on the strong convexity and smoothness of the loss). I would argue that if models would be initialized this way, then performing a clustering on the initial models should already give the right clusters. A minor issue is that the convergence rate seems not to address the number of participating workers (line 4 of Algo.