Unsupervised automated hierarchical data clustering based on simulation of a similarity matrix evolution

Abstract

A method for automated hierarchical clustering as a result of simulation of a similarity matrix evolution by the use of an algorithm for matrix transformation inducing a cooperative exchange of information within the entire pool of matrix components and resulting in retrieval of non-obvious information from the underlying data. Hierarchical clustering of matrix components is carried out by iterative transformation and division followed by transformation and division of each level clusters, or by monitoring the changes in individual binary similarities throughout transformation of the entire matrix. For monitoring of changes occurring upon matrix evolution, matrix attenuation technique is applied through the use of the mechanism of contrasting, which permits to pinpoint similarity value changes within a range of thousandths of percent. A matrix undergoing the evolution by the effect of the algorithm for matrix transformation acts as a self-learning system capable for accurate allocation of produced and accumulated information and displays abilities of a constructive element of an artificial intelligence system.

Dependencies are inferred from the claim text.