A constant-factor bi-criteria approximation guarantee for K-means++

Abstract

This paper studies the k-means++ algorithm for clustering as well as the class of Dℓ sampling algorithms to which k-means++ belongs. It is shown that for any constant factor β > 1, selecting βk cluster centers by Dℓ sampling yields a constant-factor approximation to the optimal clustering with k centers, in expectation and without conditions on the dataset. This result extends the previously known O(logk) guarantee for the case β = 1 to the constant-factor bi-criteria regime. It also improves upon an existing constant-factor bi-criteria result that holds only with constant probability.