English

Monte Carlo approximation certificates for k-means clustering

Machine Learning 2017-10-04 v1 Optimization and Control

Abstract

Efficient algorithms for kk-means clustering frequently converge to suboptimal partitions, and given a partition, it is difficult to detect kk-means optimality. In this paper, we develop an a posteriori certifier of approximate optimality for kk-means clustering. The certifier is a sub-linear Monte Carlo algorithm based on Peng and Wei's semidefinite relaxation of kk-means. In particular, solving the relaxation for small random samples of the dataset produces a high-confidence lower bound on the kk-means objective, and being sub-linear, our algorithm is faster than kk-means++ when the number of data points is large. We illustrate the performance of our algorithm with both numerical experiments and a performance guarantee: If the data points are drawn independently from any mixture of two Gaussians over Rm\mathbb{R}^m with identity covariance, then with probability 1O(1/m)1-O(1/m), our poly(m)\operatorname{poly}(m)-time algorithm produces a 3-approximation certificate with 99% confidence.

Keywords

Cite

@article{arxiv.1710.00956,
  title  = {Monte Carlo approximation certificates for k-means clustering},
  author = {Dustin G. Mixon and Soledad Villar},
  journal= {arXiv preprint arXiv:1710.00956},
  year   = {2017}
}

Comments

8 pages

R2 v1 2026-06-22T22:01:50.210Z