Machine Unlearning for Federated Socially Fair k-Means Clustering
Reddy, S K, Raza and Prakash · arXiv 2610.13717
The outcome is the verdict on the one claim below, under the protocol we fixed in advance. It is not a verdict on the paper as a whole.
Claim tested
Replay is about 35.8x faster than retraining for whole-client deletion.
What we ran and what came back
The repository's own benchmark for deleting a whole client, at a pinned commit on a 4-core cloud CPU. Retraining from scratch took 35.9 times as long as the replay path, pooled over 15 cases (interval 35.5 to 36.2; bound 18; paper 35.8). By dataset: Adult 25.8, Bank 130.7, Credit 25.1, each against a bound of 12. In every case the retrained and replayed outputs were byte-identical and matched the checksums the authors published. The 15 cases are repeats of three deletion requests, not 15 independent ones, and the ratio partly reflects the authors' choice of baseline implementation. Nothing here concerns clustering quality, other datasets or other client counts.
What this result supports
The replay speedup is real on the released frozen data at T=0.
What it does not support
Other deletion requests, T>0, other hardware.
Record
Protocol, frozen by git commit before the run and deposited afterwards (the archive was unreachable at the time): 10.5281/zenodo.23451673
The authors were sent this result. They have not reviewed this listing.
How the protocol works is on the method page. If you are an author of this paper and something here is wrong, write to customers@aegismind.app and we will correct it on this page.