|
|
Outdoor robot localization
Marek, Jakub ; Obdržálek, David (advisor) ; Iša, Jiří (referee)
The work describes the design and implementation of a framework for building control systems for mobile robots. It is a simple modular system simplifying creation of control systems and experiments with the modules. The framework contains also several tools for localization in outdoor environment.
|
|
|
Vstupní data a jejich význam pro vrstevnaté neuronové sítě
Gabašová, Evelina ; Mrázová, Iveta (advisor) ; Iša, Jiří (referee)
In the present work we study In some areas, artificial feed forward neural networks are still a competitive machine learning model. Unfortunately they tend to overfit the training data, which limits their ability to generalize. We study methods for regularization based on enforcing internal structure of the network. We analyze internal representations using a theoretical model based on information theory. Based on this study, we propose a regularizer that minimizes the overall entropy of internal representations. The entropy-based regularizer is computationally demanding and we use it primarily as a theoretical motivation. To develop an efficient and flexible implementation, we design a Gaussian mixture model of activations. In the experimental part, we compare our model with the existing work based on enforcement of internal representations. The presented Gaussian mixture model regularizer yields better results especially for classification tasks.
|
|
|
Grammatical evolution
Nohejl, Adam ; Mráz, František (advisor) ; Iša, Jiří (referee)
Grammatical evolution (GE) is a recent grammar-based approach to genetic programming that allows development of solutions in an arbitrary programming language. Its existing implementations lack documentation and do not provide reproducible results suitable for further analysis. This thesis summarises the methods of GE and the standard methods used in evolutionary algorithms, and reviews the existing implementations, foremost the only actively developed one, GEVA. A new comprehensive software framework for GE is designed and implemented based on this review. It is modular, well-documented, portable, and gives reproducible results. It has been tested in two benchmark applications, in which it showed competitive results and outperformed GEVA 10 to 29 times in computational time. It is also shown how to further improve the performance and results by using techniques unsupported by GEVA, including new modications to the previously published methods of bit-level mutation and "sensible" initialisation. The thesis and the software together form a solid foundation for further experiments and research.
|
|
|
Artificial neural networks for clustering and rule extraction
Iša, Jiří ; Mrázová, Iveta (advisor) ; Jiroutek, Pavel (referee)
Rule extraction with neural networks has been a common research topic over the last decades. This master thesis proposes a novel growing fuzzy inference neural network, based on the principle of growing neural structures [5]. This allows the network to adjust iteratively its number of hidden neurons. For the purpose of this network an existing clustering algorithm is enhanced to improve the sensitivity to the requested output. A novel fast weights adaptation, inspired by the fuzzy set theory, is also suggested. The characteristics of the proposed model and a new method of the selection of significant input features support the induction of a relatively small amount of simple fuzzy rules. The introduced techniques have been experimentally tested on real-world data describing the relationship between various types of housing in the Boston area and its price. The data was obtained from the "Boston housing" dataset.
|
|
|
Artificial intelligence algorithms for chess
Ivanová, Marika ; Petříček, Martin (advisor) ; Iša, Jiří (referee)
The primary objective of this thesis is to study and summarize the current algorithms used in computer chess and their subsequent implementation. The thesis includes a program that in addition to regular single-player game of chess against the computer leaves the user the option of used algorithms and setting their parameters. The theoretical part focuses on chess programming procedures and their illustrative explanation. Several experiments were performed with artificial intelligence algorithms and their results are presented here.
|
|
|
Upward view as a sensor in mobile robotics
Musílek, Jan ; Iša, Jiří (advisor) ; Krajíček, Václav (referee)
The use of a wide spectrum of sensors is an inherent aspect of mobile robotics. This work examines nontraditional use of the camera, directed vertically upwards, as a sensor for orienting in space. It describes various objects that can be detected in the camera image, focusing on so-called points of interest, and their pairing between two images. Based on the approximate solution of a system of linear equations, we provide information on the relative and absolute position of the robot. The work provides analysis of pairs of images, differing only the robot's rotation in place over various angles. The goal was to demonstrate the possibility of substitution of the electronic compass in cases of strong electromagnetic inter ference. Data from the image analysis was compared with that from the compass, allowing determination of the accuracy of the proposed method. The result is that upward view can be indeed used as a replacement for the electronic compass. Additionally, several modifications to the images are proposed. We examine impact of these modification on the accuracy of the suggested method. Finally, attention is given to the robotic platform on which the experiments were carried out, and the results of the measurements are interpreted.
|
|
|
Comparison of Localization Techniques
Skalka, Marek ; Obdržálek, David (advisor) ; Iša, Jiří (referee)
This work compares localization techniques used in mobile robotics. Localization - how to determine one's own position within a space - is one of the fundamental challenges of robotics. The introduction is devoted to a detailed description of localization and to the categorization of localization techniques. In subsequent chapters, category by category, various localization techniques and their variants are described and their strengths and weaknesses are compared. The work successively addresses: probabilistic localization techniques used for inaccurate sensor measurements processing and for providing reliable position estimate; relative localization techniques used for evaluation of relative changes in the robot position; and absolute localization techniques for finding and estimating the absolute position of the robot in the environment.
|
|
|
Self-organization and artificial neural networks for knowledge extraction
Aharkava, Larysa ; Mrázová, Iveta (advisor) ; Iša, Jiří (referee)
Neural networks are widely used for nancial time series prediction. However, the future values' prediction has its drawbacks and often cannot be converted to the e ffective and pro table trading system. In that thesis I will describe several di erent types of neural networks. Then, I will propose and evaluate on real series data two di erent approaches based on Kohonen's self-organizing maps and back propagation networks of how to use those networks for creating successful and pro table trading models. Also, I will give a general overview of the Forex market (Foreign exchange market) and neural networks' usage within that market.
|
|
|
Monte Carlo Localization on a graph
Pilát, Ondřej ; Obdržálek, David (advisor) ; Iša, Jiří (referee)
This bachelor thesis concerns the Monte Carlo localization on a graph. Topics researched and solved were: movement of particles on the graph, generating particles on the edge of the graph according to the last measurements, correction from localization errors, evaluation of particles based on a different types of sensors and creation of the graph from a freely available map. The utilization of the graph as a map of environment and appropriately designed movement of particles on the graph make global localization of an autonomous robot in the environment easier. The implementation was tested on a real set of data from several autonomous robots. As a result, it was observed that movement of particles on edges of the graph well represents the movement of the robot on real paths and ensures high resistancy against senzor errors.
|
|
|
Grammar-based genetic programming
Nohejl, Adam ; Mráz, František (advisor) ; Iša, Jiří (referee)
Tree-based genetic programming (GP) has several known shortcomings: difficult adaptability to specific programming languages and environments, the problem of closure and multiple types, and the problem of declarative representation of knowledge. Most of the methods that try to solve these problems are based on formal grammars. The precise effect of their distinctive features is often difficult to analyse and a good comparison of performance in specific problems is missing. This thesis reviews three grammar-based methods: context-free grammar genetic programming (CFG-GP), including its variant GPHH recently applied to exam timetabling, grammatical evolution (GE), and LOGENPRO, it discusses how they solve the problems encountered by GP, and compares them in a series of experiments in six applications using success rates and derivation tree characteristics. The thesis demonstrates that neither GE nor LOGENPRO provide a substantial advantage over CFG-GP in any of the experiments, and analyses the differences between the effects of operators used in CFG-GP and GE. It also presents results from a highly efficient implementation of CFG-GP and GE.
|
guest ::
