Modul pro Q-Learning pro Modeler neuronových sítí

Abstract

This thesis focuses on extension of Neural net modeler by module for one of the reinforcement learning algorithm called Q-learning with the use of artificial neural network. The aim is to demonstrate this type of learning on chosen environment types of classic arcade videogames from Atari company. Thesis is also focused on parallelization of this method of learning. In this thesis are described assumptions to understand problematics connection of Q-learning algorithm with artificial neural networks and its adaptation on assigned type of testing environment. Then the design and developing process of this thesis task, algorithm parameters setting methods, technologies used for implementation and conclusion evaluating the result.