Geometric Affinity Propagation for Clustering With Network Knowledge

Clustering data into meaningful subsets is a major task in scientific analysis. To date, various strategies ranging from model-based approaches to data-driven schemes, have been devised for efficient and accurate clustering. One important class of clustering methods that particular interest the exemplar-based approaches. This primarily stems amount compressed information encoded these exemplars effectively reflect characteristics corresponding clusters. Affinity propagation (AP) has proven be powerful approach refines set optimal by iterative pairwise message updates. However, critical limitation its inability capitalize on known networked relations between points often available datasets. address this shortcoming, we propose Geometric-AP, novel algorithm extends original AP take advantage network topology. Geometric-AP obeys constraints uses max-sum belief leverage topology generating smooth clusters over network. Extensive performance assessment shows leads significant quality enhancement results when compared existing schemes. Especially, demonstrate performs extremely well even cases where fails drastically.