English

A Faster $k$-means++ Algorithm

Data Structures and Algorithms 2024-02-15 v2 Machine Learning

Abstract

kk-means++ is an important algorithm for choosing initial cluster centers for the kk-means clustering algorithm. In this work, we present a new algorithm that can solve the kk-means++ problem with nearly optimal running time. Given nn data points in Rd\mathbb{R}^d, the current state-of-the-art algorithm runs in O~(k)\widetilde{O}(k ) iterations, and each iteration takes O~(ndk)\widetilde{O}(nd k) time. The overall running time is thus O~(ndk2)\widetilde{O}(n d k^2). We propose a new algorithm \textsc{FastKmeans++} that only takes in O~(nd+nk2)\widetilde{O}(nd + nk^2) time, in total.

Keywords

Cite

@article{arxiv.2211.15118,
  title  = {A Faster $k$-means++ Algorithm},
  author = {Jiehao Liang and Somdeb Sarkhel and Zhao Song and Chenbo Yin and Junze Yin and Danyang Zhuo},
  journal= {arXiv preprint arXiv:2211.15118},
  year   = {2024}
}
R2 v1 2026-06-28T07:14:31.432Z