Unit:
Κατεύθυνση Ηλεκτρονικός Αυτοματισμός (Η/Α, με πρόσθετη εξειδίκευση στην Πληροφορική και στα πληροφοριακά συστήματα)Library of the School of Science
Author:
Mavrothalassitis Kyriakos
Supervisors info:
Διονύσιος Ρεΐσης, Αναπληρωτής Καθηγητής, Τμήμα Φυσικής, ΕΚΠΑ
Έκτορας Νισταζάκης, Αναπληρωτής Καθηγητής, Τμήμα Φυσικής, ΕΚΠΑ
Δρ Νικόλαος Βλασσόπουλος, Επιστημονικός Συνεργάτης
Original Title:
Artificial Intelligence with Reinforcement Learning on Video-Games
Translated title:
Artificial Intelligence with Reinforcement Learning on Video-Games
Summary:
The purpose of this thesis was the implementation of an Artificial Intelligence (AI) system via software so that it would learn to play video games.
In the first chapter the three basic types of Machine Learning (ML) were discussed in short and the type of Machine Learning that is related to our problem was specified.
In chapter two, the theory on which Neural Networks (NN) are based was reviewed in detail. In particular we referred to the neurons of the human brain, the Perceptron and Adaline artificial neurons and the types of Neural Networks that can be constructed by modern artificial neurons. Finally, Convolutional Neural Networks (CNN) were detailed, as well as the Long Short-Term Memory Neural Networks (LSTM) that were used in this thesis extensively.
In chapter three, we studied the theory of Reinforcement Learning (RL). Particularly, Markov Decision Processes (MDP), Bellman equations, the type of systems which operate with Markov Decision Processes and the way of training systems of Reinforcement Learning were discussed. Finally, Google’s Asynchronous Advantage Actor-Critic (A3C) algorithm for Reinforcement Learning, used for the development of the software in this thesis, was presented in detail.
In chapter four, we presented the game used for training the software of this thesis. Furthermore, the types of Neural Networks which were used as components for the implementation of our system of Artificial Intelligence were analyzed.
In the fifth and final chapter, the experiments and the results of the training process of the system were presented.
Main subject category:
Science
Keywords:
artificial, intelligence, reinforcement, learning, video-games
