Modul samoorganizačních map pro program Modeler neuronových sítí

Abstract

This diploma thesis deals with the extension of the Neural Net Modeler program by a module of self-organizing maps with the possibility of choosing different types of neighborhoods. This program is used in teaching the subject Neural Networks. The module allows you to create, teach and visualize the internal structure of these networks. A program for parallel running of self-organizing maps on distributed computing nodes was also implemented. The work explains the basics of neural networks with a focus on self-organizing maps. The next chapter describes the issue of parallelization of the learning process of these networks. One chapter is devoted to the datasets used in the testing of neural networks. The following chapters describe the implemented module and the program for parallelization of learning. The last chapter analyzes and compares the various methods of parallelization and the influence of the neighborhood on the learning process.