Piecewise Linear Manifold Clustering

Clustering is commonly viewed as an ill-defined problem because there is not clear performance criteria which can tell about relevant element arrangements withing the clustering. Devising such evaluation criteria usually based on geometrical and/or structural composition of the original data. We present a novel algorithm that incorporates a topological structure and information-theoretical description of the clustered data to recognize and evaluate stable nonlinear structures in form of a piecewise linear manifolds as a meta-extension to common statistical clustering algorithms.