Reinforcement Learning in Machine Learning

Cilj je ovoga diplomskog rada objasniti učenje podrškom - paradigmu učenja neuronskih mreža koja se zasniva na metodi pokušaja i pogrešaka. U tu je svrhu prvo opisan način rada neuronskih mreža te osnovni elementi istih. Definiran je i Markovljev proces odlučivanja koji je osnova algoritama Q-učenja koji se koriste za rješavanje problema metodom učenja podrškom te su opisani algoritmi Q-učenja s iteracijama vrijednosti i dubinskog Q-učenja. Na kraju je dan primjer implementacije dubinskog Q-učenja u programskom jeziku Python.The goal of this master’s thesis is to explain reinforcement learning - neural network learning method based on trial and error principle. For that purpose, the way that neural networks work and their basic elements are first described. Furthermore, Markov decision process, which is the basis of Q-learning algorithms that are being used to solve problems with reinforcement learning, is defined and algorithms of Q-learning with value iterations and deep Q-learning are described. An example of implementation of deep Q-learning written in Python programming language is given at the end