|
|
Genetic Algorithms driven by MCTS
Havránek, Štěpán ; Hric, Jan (advisor) ; Moudřík, Josef (referee)
Evolutionary and genetic algorithms are problem-solving methods designed according to a nature inspiration. They are used for solving hard problems that we cannot solve by any efficient specialized algorithm. The Monte Carlo method and its derivation the Monte Carlo Tree Search (MCTS) are based on sampling and are also commonly used for too complex problems, where we are dealing with enormous memory consumption and it is impossible to perform a complete searching. The goal of this thesis is to design a general problem solving method that is built from these two completely different approaches. We explain and implement the new method on one example problem: the Traveling salesman problem (TSP). Second part of this thesis contains various tests and experiments. We compare different settings and parametrizations of our method. The best performing variant is then compared with the classical evolutionary TSP solution or, for example, with greedy algorithms. Our method shows competitive results. The best results were achieved with the cooperation of our method and the classical evolutionary TSP solution. This union shows better results than any of its parts separately, which we find as a great success.
|
|
|
Detection and analysis of polychronous groups emerging in spiking neural network models.
Šťastný, Bořek ; Brom, Cyril (advisor) ; Moudřík, Josef (referee)
How is information represented in real neural networks? Experimental results continue to provide evidence for presence of spiking patterns in network activity. The concept of polychronous groups attempts to explain these results by proposing that neurons group together to fire in non- synchronous but precise time-locked chains. Several methods for the detection of such groups have been proposed, however, they all employ extensive searching in network structure, which limits their usefulness. We present a new method by observing spiking dependencies in network activity to directly detect polychronous groups. Our method shows comparatively more efficient computation by trading off detection selectivity. The method allows for analysis of polychronous groups emerging in noisy networks. Our results support the existence of structure-forming properties of spontaneous activity in neural network.
|
|
|
Model V1 s realistickou distribucí funkčních typů neuronů v rámci kortikálních vrstev
Moudřík, Josef ; Brom, Cyril (advisor) ; Sýkora, Ondřej (referee)
There have been identified two functionally different neuron classes in the primary visual cortex (V1), so-called simple and complex cells. These cells differ in reactions to various stimuli and their development has been successfully simulated in one computational model of V1. This model, however, simulates both classes in separate layers corresponding to layers 4C and 2/3 in V1. On the contrary, experiments have shown that both categories are - in different proportions - present in both layers. In this thesis, a computational model with a realistic distribution of complex and simple cells is presented. To increase its authenticity, I incorporate long-range excitatory and short-range inhibitory lateral cortical connections as found in animals, overcoming one drawback of previous models that used long-range inhibition. To assess my model, two measures of orientation selectivity - circular variance and orientation bandwidth - were computed for each simulated neuron. Using this measures, I compared my model with data from macaque monkey. In line with biological findings, my model develops a wide diversity of orientation selectivity. Moreover, it develops maps of orientation preference and realistic receptive fields.
|
|
|
Dance Recognition from Audio Recordings
Pavlín, Tomáš ; Čech, Jan (advisor) ; Moudřík, Josef (referee)
We propose a CNN-based approach to classify ten genres of ballroom dances given audio recordings, five latin and five standard, namely Cha Cha Cha, Jive, Paso Doble, Rumba, Samba, Quickstep, Slow Foxtrot, Slow Waltz, Tango and Viennese Waltz. We utilize a spectrogram of an audio signal and we treat it as an image that is an input of the CNN. The classification is performed independently by 5-seconds spectrogram segments in sliding window fashion and the results are then aggregated. The method was tested on following datasets: Publicly available Extended Ballroom dataset collected by Marchand and Peeters, 2016 and two YouTube datasets collected by us, one in studio quality and the other, more challenging, recorded on mobile phones. The method achieved accuracy 93.9%, 96.7% and 89.8% respectively. The method runs in real-time. We implemented a web application to demonstrate the proposed method.
|
|
|
Detection and analysis of polychronous groups emerging in spiking neural network models.
Šťastný, Bořek ; Brom, Cyril (advisor) ; Moudřík, Josef (referee)
How is information represented in real neural networks? Experimental results continue to provide evidence for presence of spiking patterns in network activity. The concept of polychronous groups attempts to explain these results by proposing that neurons group together to fire in non- synchronous but precise time-locked chains. Several methods for the detection of such groups have been proposed, however, they all employ extensive searching in network structure, which limits their usefulness. We present a new method by observing spiking dependencies in network activity to directly detect polychronous groups. Our method shows comparatively more efficient computation by trading off detection selectivity. The method allows for analysis of polychronous groups emerging in noisy networks. Our results support the existence of structure-forming properties of spontaneous activity in neural network.
|
|
|
General Artificial Intelligence for Game Playing
Klůj, Jan ; Pilát, Martin (advisor) ; Moudřík, Josef (referee)
Game playing is a relatively interesting task in the field of artificial intelligence in these days. The master thesis deals with general artificial intelligence which is capable of playing selected simple games based on information that is also avai- lable to the human player. Our selected games are 2048, Mario, racing simulator TORCS and Alhambra. All the information acquired by artificial intelligence is provided by games through an interface, therefore none of the models uses visual input. We use evolutionary approaches such as evolutionary algorithms, evolutio- nary strategy CMA and differential evolution applied to different types of neural networks. We are also dealing with deep reinforcement learning. We test these approaches and compare their results. 1
|
|
|
HexMage - Encounter Balancing in Hex Arena
Arnold, Jakub ; Gemrot, Jakub (advisor) ; Moudřík, Josef (referee)
Title: HexMage - Encounter Balancing in Hex Arena Author: Jakub Arnold Department: Department of Software and Computer Science Education Supervisor: Mgr. Jakub Gemrot, Department of Software and Computer Science Education Abstract: Procedural content generation (PCG) is mostly examined in the con- text of map/environment creation, rather than generating the actual game char- acters. The goal of this thesis is to design a turn-based RPG-like game with perfect information for which we can generate balanced encounters. The game consists of a hex-based arena in which two teams fight. Each team consists of a few player controller characters with unique abilities. We generate the attributes of these abilities in order to make the encounter balanced. We will also build an AI that can be used to automatically play-test the PCG algorithm. The goal is to generate an equally strong, but different opponent. Keywords: video games encounter balancing hex arena rpg elements 1
|
|
|
Genetic Algorithms driven by MCTS
Havránek, Štěpán ; Hric, Jan (advisor) ; Moudřík, Josef (referee)
Evolutionary and genetic algorithms are problem-solving methods designed according to a nature inspiration. They are used for solving hard problems that we cannot solve by any efficient specialized algorithm. The Monte Carlo method and its derivation the Monte Carlo Tree Search (MCTS) are based on sampling and are also commonly used for too complex problems, where we are dealing with enormous memory consumption and it is impossible to perform a complete searching. The goal of this thesis is to design a general problem solving method that is built from these two completely different approaches. We explain and implement the new method on one example problem: the Traveling salesman problem (TSP). Second part of this thesis contains various tests and experiments. We compare different settings and parametrizations of our method. The best performing variant is then compared with the classical evolutionary TSP solution or, for example, with greedy algorithms. Our method shows competitive results. The best results were achieved with the cooperation of our method and the classical evolutionary TSP solution. This union shows better results than any of its parts separately, which we find as a great success.
|
|
|
Meta-learning methods for analyzing Go playing trends
Moudřík, Josef ; Neruda, Roman (advisor) ; Mráz, František (referee)
This thesis extends the methodology for extracting evaluations of players from samples of Go game records originally presented in (Baudiš - Moudřík, 2012). Firstly, this work adds more features and lays out a methodology for their comparison. Secondly, we develop a robust machine-learning framework, which is able to capture dependencies between the evaluations and general target variable using ensemble meta-learning with a genetic algorithm. We apply this framework to two domains, estimation of strength and styles. The results show that the inference of the target variables in both cases is viable and reasonably precise. Finally, we present a web application, which realizes the methodology, while presenting a prototype teaching aid for the Go players and gathering more data. Powered by TCPDF (www.tcpdf.org)
|
|
|
Model V1 s realistickou distribucí funkčních typů neuronů v rámci kortikálních vrstev
Moudřík, Josef ; Sýkora, Ondřej (referee) ; Brom, Cyril (advisor)
There have been identified two functionally different neuron classes in the primary visual cortex (V1), so-called simple and complex cells. These cells differ in reactions to various stimuli and their development has been successfully simulated in one computational model of V1. This model, however, simulates both classes in separate layers corresponding to layers 4C and 2/3 in V1. On the contrary, experiments have shown that both categories are - in different proportions - present in both layers. In this thesis, a computational model with a realistic distribution of complex and simple cells is presented. To increase its authenticity, I incorporate long-range excitatory and short-range inhibitory lateral cortical connections as found in animals, overcoming one drawback of previous models that used long-range inhibition. To assess my model, two measures of orientation selectivity - circular variance and orientation bandwidth - were computed for each simulated neuron. Using this measures, I compared my model with data from macaque monkey. In line with biological findings, my model develops a wide diversity of orientation selectivity. Moreover, it develops maps of orientation preference and realistic receptive fields.
|
guest ::
RSS feed.