Modul LSTM a Rekurentních neuronových sítí pro program Modeler neuronových sítí

Abstract

This master thesis focuses on the implementation of an extension module for the Neuron Net Modeler program that is used as a support for Neural Networks subject. This module allows you to create, configure, teach and test recurrent neural networks without their in-depth knowledge. The thesis furthermore explains the basic concepts and principles of neural networks, in detail focused on recurrent networks. Prototype parallelization of learning of these networks for learning on computing nodes using data parallelism is also part of the thesis. The conclusion is devoted to verifying the correct implementation and performing a series of experiments with both sequential and parallel learning of recurrent neural networks.