NeurIPS 2020

An Efficient Framework for Clustered Federated Learning

Meta Review

Reviewers agree that the central idea is simple, which can be seen as a strength, and that the analysis is valuable. The concern about comparison only to baselines and not a more real-world method will be rectified by including the promised comparison to ClusteredFL. Without this comparison at submission, we must assume it will be on par, and therefore the significance of the result is reduced. The statements about reduced computation at the central server can also be accompanied by the statements abour privacy benefits (not sending user data to the server), even given the provisos at line 347.