Towards unsupervised data-flow analysis: neural models for clustering and factor analysis of large sets of highly multidimensional objects

摘要

Two stochastic neural models implementing a mix of clustering and factor analysis techniques are presented: the axial k-means and a more sophisticated local component analysis. Both converge to a local (resp. global) optimum of their objective function. Simulations and comparisons with classical algorithms are presented. The dynamicity of the model, i.e. instantaneous adaptation to any new data vector, is a desirable feature if many applications,.