|
|
Evolutionary Algorithms for Data Transformation
Švec, Ondřej ; Pilát, Martin (advisor) ; Neruda, Roman (referee)
In this work, we propose a novel method for a supervised dimensionality reduc- tion, which learns weights of a neural network using an evolutionary algorithm, CMA-ES, optimising the success rate of the k-NN classifier. If no activation func- tions are used in the neural network, the algorithm essentially performs a linear transformation, which can also be used inside of the Mahalanobis distance. There- fore our method can be considered to be a metric learning algorithm. By adding activations to the neural network, the algorithm can learn non-linear transfor- mations as well. We consider reductions to low-dimensional spaces, which are useful for data visualisation, and demonstrate that the resulting projections pro- vide better performance than other dimensionality reduction techniques and also that the visualisations provide better distinctions between the classes in the data thanks to the locality of the k-NN classifier. 1
|
|
|
Placement of map symbols
Burešová, Karolína ; Mareš, Martin (advisor) ; Pangrác, Ondřej (referee)
Placing map symbols in a way so that the resulting map looks well is a major problem in cartography. In this thesis, we deal with automatization of this process using an evolutionary algorithm. Input of this algorithm is a set of requests for symbols (signposts, ruins, street names etc) to be placed on the map, its output is a description of their placement. Unlike other studies, we deal with labels for all kind of features (not only point-features) as well as with placing the features themselves. We managed to design an evolutionary algorithm which produces acceptable maps and offers some possibilities to further enhance the quality of produced maps.
|
|
| |
|
|
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.
|
|
|
Genetic Programming for Control of Robotic Swarms
Filippi, Michal ; Pilát, Martin (advisor) ; Děchtěrenko, Filip (referee)
Homogeneous robotic swarms are usually controlled by a manually created program. This thesis studies an alternative approach, the possibilities of creating control programs by means of a technique inspired by biological evolution called genetic programming. A simulator of a simple 2D environment was created for this purpose. This allows us to observe and examine newly created control programs for virtual homogeneous robotic swarm. The ability of genetic programming to create control programs is examined on three different scenarios in which the robotic swarm should deal with three different tasks. The thesis also contains the comparison of genetic programming with a technique that use neural network and evolutionary strategies. Powered by TCPDF (www.tcpdf.org)
|
|
|
Koza and Prolog
Frauknecht, Jan ; Švarný, Petr (advisor) ; Verner, Jonathan (referee)
This paper introduces the artificial intelligence background of ge- netic programming and some properties of logical programming para- digm. However, the main task of this work is to create the genetic programming algorithm that operates with the logical programming paradigm. SWI-Prolog is used for the actual implementation of such a program. This implementation is in detail described. The testing of this implementation shows some possible path for the future work. 1
|
|
|
Evolution of behaviors for intelligent agents
Obrázek, Václav ; Neruda, Roman (advisor) ; Surynek, Pavel (referee)
This thesis deals with agent behavior evolution for the environment of a real computer game using evolutionary algorithms. The game Unreal Tournament 2004 was chosen, due to its ease of use for creating agents manually with the Pogamut suite of tools. As a decision making structure for the agents yaPOSH reactive plans were used. Due to the demanding needs on the hardware and time a real computer game is not considered to be very suitable for artificial evolution. To overcome this fact a light-weighted environment LightEnv, that simulates only those aspects that are important for agent evolution, was created. The evolution was based on genetic programming modified for use with yaPOSH reactive plans. The evolved agent behavior for death match and team death match game scenarios exceeded the preprogrammed ones and was successfully transferred to Unreal Tournament 2004 environment. In the team death match scenario an interesting behavior that utilizes agent communication was evolved.
|
|
|
Artificial intelligence for the game Desktop dungeons
Černý, Vojtěch ; Děchtěrenko, Filip (advisor) ; Pilát, Martin (referee)
Rogue-like games are a subgenre of computer RPG games, featuring procedurally generated environment and permanent death. Winning them is a challenge for a human player, and more so for artificial intelligence (AI). In this work, we present a framework for implementing artificial players for a rogue-like game Desktop Dungeons. We then investigate options of suitable AI creation, and settle for using a genetic algorithm to fine-tune a greedy strategy. The resulting AI was as succesful as a mediocre human player, winning the game 72% of the time. This framework and results may be used to improve the quality of rogue-like games, procedural content generators, and artificial intelligence in similiar environments. Powered by TCPDF (www.tcpdf.org)
|
|
|
Genetic programming in financial markets forecasting
Krejčí, Tomáš ; Bednárek, David (advisor) ; Majerech, Vladan (referee)
The aim of this thesis is to test usability of the genetic programming for predicting of the financial markets based on historical prices. The thesis includes the study of genetic programming techniques used or useful for the market prediction. The practical part of thesis is implementation of selected methods and testing their performance on available historical data from financial markets. Powered by TCPDF (www.tcpdf.org)
|
|
|
Spatial modeling of brain tissue
John, Pavel ; Neruda, Roman (advisor) ; Brom, Cyril (referee)
Neural connections in the human brain are known to be modified by experiences. Yet, little is known about the mechanism of the modification and its implications on the brain function. The aim of this thesis is to investigate what impact the spatial properties of brain tissue can have on learning and memory. In particular, we focus on the dendritic plasticity. We present a model where the tissue is represented by a two-dimensional grid and its structure is characterized by various connections between the grid cells. We provide a formal definition of the model and we prove it to be computational as strong as the Turing machine. An adaptation algorithm proposed enables the model to reflect the environmental feedback, while evolutionary algorithms are employed to search for a satisfactory architecture of the model. Implementation is provided and several experiments are driven to demonstrate the key properties of the model. Powered by TCPDF (www.tcpdf.org)
|
guest ::
