Machine Unlearning for Federated Socially Fair k-Means Clustering

Hospitals that cluster patients jointly, without pooling records, must treat demographic groups fairly and honour deletion requests. We build k-means++ summaries for each institution and group, prove an O(log² k) guarantee that does not depend on how groups are split across institutions, and show that replay reproduces retraining exactly and 35.8× faster when an institution withdraws. The guarantee uses the k-means++ bound of our NeurIPS 2020 paper as a black box.