Adaptive Affinity Propagation clustering

advantage of speed & performance appears under large number of clusters & large dataset

Affinity propagation clustering (AP) is a clustering algorithm proposed in "Brendan J. Frey and Delbert Dueck. Clustering by Passing Messages Between Data Points. Science 315, 972 (2007)". It has some advantages: speed, general applicability, and suitable for large number of clusters. AP has two limitations: it is hard to known what value of parameter ‘preference’ can yield optimal clustering solutions, and oscillations cannot be eliminated automatically if occur.

Adaptive AP improves AP in these items: adaptive adjustment of the damping factor to eliminate oscillations (called adaptive damping), adaptive escaping oscillations, and adaptive searching the space of preference parameter to find out the optimal clustering solution suitable to a data set (called adaptive preference scanning). With these adaptive techniques, adaptive AP will outperform AP algorithm in clustering quality and oscillation elimination, and it will find optimal clustering solutions by Silhouette indices.