|
Demonstrational Program for IZU Course
Hreha, Tomáš ; Šůstek, Martin (referee) ; Zbořil, František (advisor)
This bachelor thesis deals with the design of application for visualization of fundamental algorithms of artificial intelligence. The first part describes theoretical part of implemented topics and methods, next part briefly describes used technologies, reasons why they were used and their practical usage in context of result application. The next part is dedicated to user interface, its main components and describes ways how application interacts with user and how user can interact with application. The last part contains comparison with original demo applications and summarize results of application testing.
|
|
Deterministic Games Playing with Learning
Knoflíček, Jakub ; Rozman, Jaroslav (referee) ; Zbořil, František (advisor)
This paper deals with creating artificial intelligence for computer player for deterministic games such as checkers. It's analyzed principle of idea and method for best move searching for a current game state in very large state space in combination with method reinforcement learning which allows us to evaluate individual game states. The paper also involves analysis of method for finding all possible moves in concrete implemented game checkers, concept of effective treatment with evaluated game states and mechanism alternate evaluating in case the absence any of them. At the end is final application going in collection of tests where is compared with competitive program and the achieved properties of methods searching best move and reinforcement learning are analyzed.
|
|
Deep Neural Networks for Reinforcement Learning
Ludvík, Tomáš ; Bambušek, Daniel (referee) ; Hradiš, Michal (advisor)
The aim of this thesis is to use deep neural networks for task in reinforcement learning. I use my modification of 2D game Tuxánci for the purposes of the test environment. This modification provides the possibility of using the game as an environment for machine learning. Subsequently, Iam solving the task of learning the agent by using reinforcement learning with the Double DQN algorithm.
|
|
Shared Experience in Reinforcement Learning
Mojžíš, Radek ; Šůstek, Martin (referee) ; Hradiš, Michal (advisor)
The aim of this thesis is to use methods of transfer learning for training neural network on a reinforcement learning tasks. As test environment, I am using old 2D console games, such as space invaders or phoenix. I am testing the impact of re-purposing already trained models for different environments. Next I use methods for domain feature transfer. Lastly i focus on the topic of multi-task learning. From the results we can gain insight into possibilities of using transfer learning for reinforcement learning algorithms.
|
| |
|
Reinforcement Learning for Mobile Robots
Hás, David ; Zemčík, Pavel (referee) ; Hradiš, Michal (advisor)
This paper is concerned with reinforcement learning for robotic movement in simulated physical environment. These are difficult problems for reinforcement learning, where agents need to face several challenges. One of them is continuous action space, as agent usually interacts with the environment by applying force on joints of the robot. Another problem is that parts of the robot often affect each other in complex ways and are also affected by gravity, inertia and other physical effects. For these and more reasons simple reinforcement learning algorithms are not suitable for these tasks. One of recent solutions is the Soft Actor-Critic algorithm (SAC), which emerged at the same time as similarly performing TD3, and both outperforming the older DDPG. SAC agents are rewarded for behaving more randomly, thus their goal is to maximize entropy along with maximizing the reward. This paper describes usage of this algorithm for teaching agents robotic movement. It describes implementation of the algorithms using the PyTorch machine learning framework and evaluates it on environments from OpenAI Gym platform using the PyBullet physics engine. Lastly, the algorithm is applied on custom environment with robot Atlas.
|
|
Deep Neural Networks for Reinforcement Learning
Košák, Václav ; Dobeš, Petr (referee) ; Hradiš, Michal (advisor)
The paper describes a training environment for training a character how to walk. The environment is implemented in Al Gym by using the PyBullet physical model. Tasks from the environment are solved by using reinforcement learning by the ActorCritic algorithm. Each of the tasks is focused on the fundamental movements of the character. The paper show, which reward functions are used by the algorithm to solve given tasks.
|
|
Intelligent Reactive Agent for the Game Ms.Pacman
Bložoňová, Barbora ; Zbořil, František (referee) ; Drahanský, Martin (advisor)
This thesis focuses on artificial intelligence for difficult decision problemes such as the game with uncertainty Ms. Pacman. The aim of this work is to design and implement intelligent reactive agent using a method from the field of reinforcement learning, demonstrate it on visual demo Ms.Pacman and compare its intelligence with well-known informed methods of playing games (Minimax, AlfaBeta Pruning, Expectimax). The thesis is primarily structured into two parts. The theoretical part deals with adversarial search (in games), reactivity of agent and possibilities of machine learning, all in the context of Ms. Pacman. The second part addresses the design of agent's versions behaviour implementation and finally its comparison to other methods of adversarial search problem, evaluation of results and a few ideas for future improvements.
|
|
Analysis of Various Approaches to Solving Optimization Tasks
Knoflíček, Jakub ; Samek, Jan (referee) ; Zbořil, František (advisor)
This paper deals with various approaches to solving optimization tasks. In prolog some examples from real life that show the application of optimization methods are given. Then term optimization task is defined and introducing of term fitness function which is common to all optimization methods follows. After that approaches by particle swarm optimization, ant colony optimization, simulated annealing, genetic algorithms and reinforcement learning are theoretically discussed. For testing we are using two discrete (multiple knapsack problem and set cover problem) and two continuous tasks (searching for global minimum of Ackley's and Rastrigin's function) which are presented in next chapter. Description of implementation details follows. For example description of solution representation or how current solutions are changed. Finally, results of measurements are presented. They show optimal settings for parameters of given optimization methods considering test tasks. In the end are given test tasks, which will be used for finding optimal settings of given approaches.
|
|
Playing of Nondeterministic Games with Learning
Bukovský, Marek ; Rozman, Jaroslav (referee) ; Zbořil, František (advisor)
The thesis is dedicated to the study and implementation of methods used for learning from the course of playing. The chosen game for this thesis is Backgammon. The algorithm used for training neural networks is called the temporal difference learning with use of eligible traces. This algorithm is also known as TD(lambda). The theoretical part describes algorithms for playing games without learning, introduction to reinforcement learning, temporal difference learning and introduction to artificial neural networks. The practical part deals with application of combination of neural networks and TD(lambda) algorithms.
|