K-means is a popular clustering algorithm because of its simplicity and ease of implementation. But it has some big problems like the dependence of initial centroid choices for convergence time and result quality.

Hélio Almeida who is a new Post-Doc in our research team, did a nice presentation on k-means and methods to improve the “vanilla” version of this method, like “k-means plus plus” and “k-means parallel”. This presentation is available by clicking on this link.