Data mining processing based on problem-oriented machine architecture

Such tasks as clustering, classification, regression, and associative search are viewed as formal tasks Data Mining. Objects presented in the form of multi-dimensional numerical vectors of informative features are initial data. For each of the tasks exist many known algorithms of solving, but they have fundamental differences. The first group is based on calculation of the distance between the images in a specific metric, the second group is based on approximation of regression or discriminate function, and others are based on analysis and processing of math graphs. These differences are a cause for very narrow application of known neuro-computers and associative processors. At the same time, the universal parallel processors (multi-core CPU and GPU) have objective limitations in accordance with the Amdahl's and Gustafson's Laws, and aren't optimal for solving Data Mining. We propose an approach that unifies hardware and software implementation of the tasks of regression, classification, clustering and associative search through the unification of different algorithms in a unified mathematical model.