Numerical acceleration of data processing using MATLAB for the needs of expert systems

Abstract

The article deals with alternative techniques of accelerating numerical computations. Working with sets of local models requires fast processing of data in form of matrices. These data objects consist of input matrices with different dimensions and their mutual multiplication, division or other basic mathematical operations. The most suitable accelerating technique for these operations is parallelized computation. The technique can be applied in the form of using a large number of simple mathematical coprocessors which are found in modern graphics cards, so called general-purpose computation on graphics processing units, or connecting multiple computers to a distributed computing network called high performance computing cluster. The techniques are later discussed in the article. The article also contains practical comparison of both techniques and evaluation of the possibilities of their application.