K-bMOM

K-bMOM is a clustering algorithm that manages to give relevant clusterings while data are polluted with isolated datapoints (not visible on the picture).

You can see on the left that their algorithm K-bMOM (bottom right clustering) does much better than K-medians, K-means and the less known K-PDTM.

We all hope their work will be published in the prestigious Journal of American Statistical Association (JASA). Until the verdict of the journal JASA, you can find the current version of there article on arXiv (or directly here).