Using quantization techniques, Laloë (2010) defined a new clustering algorithm called Alter. This L1-based algorithm is shown to be convergent but suffers two major flaws. The number of clusters, K, must be supplied by the user and the computational cost is high. This article adapts the X-means algorithm (Pelleg & Moore, 2000) to solve both problems.
Laloë, Thomas and Servien, Rémi
"The X-Alter Algorithm: A Parameter-Free Method of Unsupervised Clustering,"
Journal of Modern Applied Statistical Methods:
1, Article 14.
Available at: http://digitalcommons.wayne.edu/jmasm/vol12/iss1/14