Papers
Topics
Authors
Recent
Search
2000 character limit reached

Further heuristics for kk-means: The merge-and-split heuristic and the (k,l)(k,l)-means

Published 23 Jun 2014 in cs.LG, cs.CV, cs.IR, and stat.ML | (1406.6314v1)

Abstract: Finding the optimal kk-means clustering is NP-hard in general and many heuristics have been designed for minimizing monotonically the kk-means objective. We first show how to extend Lloyd's batched relocation heuristic and Hartigan's single-point relocation heuristic to take into account empty-cluster and single-point cluster events, respectively. Those events tend to increasingly occur when kk or dd increases, or when performing several restarts. First, we show that those special events are a blessing because they allow to partially re-seed some cluster centers while further minimizing the kk-means objective function. Second, we describe a novel heuristic, merge-and-split kk-means, that consists in merging two clusters and splitting this merged cluster again with two new centers provided it improves the kk-means objective. This novel heuristic can improve Hartigan's kk-means when it has converged to a local minimum. We show empirically that this merge-and-split kk-means improves over the Hartigan's heuristic which is the {\em de facto} method of choice. Finally, we propose the (k,l)(k,l)-means objective that generalizes the kk-means objective by associating the data points to their ll closest cluster centers, and show how to either directly convert or iteratively relax the (k,l)(k,l)-means into a kk-means in order to reach better local minima.

Authors (2)
Citations (4)

Summary

No one has generated a summary of this paper yet.

Paper to Video (Beta)

No one has generated a video about this paper yet.

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Continue Learning

We haven't generated follow-up questions for this paper yet.

Tweets

Sign up for free to view the 1 tweet with 20 likes about this paper.