Nonlinear approaches to automatic elicitation of distributed oscillatory clusters in adaptive self-organized system

Abstract

Chaotic neural networks find more and more applications in pattern recognition systems. However hybrid multidisciplinary solutions that combine advances from physics and artificial intelligence fields tend to enrich the complexity of designed systems and add more discussion points. This paper questions the applicability of well known chaotic time-series metrics (Shannon entropy, Kolmogorov entropy, Fractal dimension, Gumenyuk metric, complete and lag synchronization estimations) to simplify elicitation of distributed oscillatory clusters that store clustering results of a problem. Computer modeling results gives evidence that in case of clustering simple datasets the metrics are rather effective; however the concept of averaging out agent's dynamics fails when the clusters in the input dataset are linearly non-separable. © 2012 Springer-Verlag.