Εξομοίωση και σύγκριση αλγορίθμων ενισχυτικής μάθησης
Simulation and comparison of reinforcement learning algorithms
View/
Keywords
Ενισχυμένη Μάθηση ; NEAT ; Deep Q ; Q learning ; Reinforcement learning ; PythonAbstract
The field of Artificial intelligence in modern era is an important tool in the implementation of modern ideas and the development of many useful applications. Machine learning provides a variety of techniques and algorithms for problem solving where classical programming is limited. A similar problem is presented in this work, which is a theoretical and practical approach in the field of machine learning aiming in finding suitable techniques for solving complex problems.
Creating intelligent artificial agents to solve human-related problems is a major challenge for the artificial intelligence community. Particularly important is the understanding of the dynamic environment in which they operate and their interaction with it, just like humans do in the physical world. The field that specializes in creating such agents is called Reinforcement Learning. With the term Reinforcement Learning, we refer to the methods through which a system of algorithms 'learns' to interact within a structured environment after trial and error. This learning is done by exploring the environment through actions and rewards given by the environment to achieve a goal with optimal effort.
This research was developed on the idea of the simple chase game, which aims to create an environment and the agents that play in it. In this game, hunting agents are tasked with catching opponents, while the hunted agents try to stay free. The goal of the work is, apart from understanding the rules and getting the right decisions to achieve victory, finding new strategies that will lead to optimal results and comparing the most valuable algorithms.
The proposed approach has been chosen after analysing a wide range of Reinforcement Learning algorithms in a graphical emulation environment. The results achieved show the best implementation technique, but at the same time prospective improvements are also highlighted in the way of getting the right decisions but also in dealing with the more complex environment and rules of the game.
Postgraduate Studies Programme
Προηγμένα Συστήματα Πληροφορικής - Ανάπτυξη Λογισμικού και Τεχνητής ΝοημοσύνηςDepartment
Σχολή Τεχνολογιών Πληροφορικής και Επικοινωνιών. Τμήμα ΠληροφορικήςNumber of pages
57Language
GreekCollections
Except where otherwise noted, this item's license is described as
Αναφορά Δημιουργού-Μη Εμπορική Χρήση 3.0 Ελλάδα
Αναφορά Δημιουργού-Μη Εμπορική Χρήση 3.0 Ελλάδα