|
|
Utilization of Ontologies for Flexible Production Modelling in Industry 4.0
Matyáš, Petr ; Jirkovský, Václav (advisor) ; Mráz, František (referee)
The topic of this Master thesis is ontologies and the methods of their design using OWL. The thesis aims to provide a manual on the design of ontologies and demonstrate the use of this manual in the environment of Industry 4.0. The actual manual is preceded by chapters presenting the used semantic web technologies from a state-of-the-art point of view. The work is concluded by the author's evaluation of the suitability of the use of given technologies. 1
|
|
|
Artificial intelligence for Sushi Go!
Filek, Jiří ; Holan, Tomáš (advisor) ; Mráz, František (referee)
The thesis deals with an artificial intelligence for a Sushi Go! card game. It is a game with simultaneous moves for two to five players. The thesis presents multiple approaches for development of an artificial intelligence. The main focus is on methods based on the MCTS algorithm, namely DUCT and EXP3. An artificial intelligence using weighted rules is tried as well. The weights are assigned by a genetic algorithm. The first part of the thesis is about game analysis and description of chosen methods. The second part of the thesis is about parameter tuning and comparison of different agents for two or more players. The comparison is based on a large number of games played between agents. The last part of the thesis deals with an implementation. A graphical and a console applications were created for the purpose of developing an artificial intelligence. The console application is used for parameter tuning and also to compare artificial intelligences. The graphical application is used for a game of human against artificial intelligence or other humans on a single computer. Overall, DUCT performs best in every experiment despite its theoretical disadvantages.
|
|
|
Machine learning on small datasets with large number of features
Beran, Jakub ; Mráz, František (advisor) ; Matzner, Filip (referee)
Machine learning models are difficult to employ in biology-related research. On the one hand, the availability of features increases as we can obtain gene expressions and other omics information. On the other hand, the number of available observations is still low due to the high costs associated with obtaining the data for a single subject. In this work we, therefore, focus on the set of problems where the number of observations is smaller than the number of features. We analyse different combinations of feature selection and classification models and we study which combinations work the best. To assess these model combinations, we introduce two simulation studies and several real-world datasets. We conclude that most classification models benefit from feature pre-selection using feature selection models. Also, we define model-based thresholds for the number of observations above which we observe increased feature selection stability and quality. Finally, we identify a relation between feature selection False Discovery Rate and stability expressed in terms of the Jaccard index. 1
|
|
|
Modeling of fragment-based molecular similarity
Lamprecht, Matyáš ; Škoda, Petr (advisor) ; Mráz, František (referee)
Virtual screening is a part of computer-aided drug design, which aims to identify biologically active molecules. The ligand-based virtual screening employs known bio- logically active molecules and similarity search. A common approach to computation of molecular similarity is to utilize molecular fingerprints. Hashed structural molecular fingerprints hash fragments (subgraphs) of molecular graphs into a bit string reducing the problem of molecular similarity to the bit string similarity. Due to the hashing two distinct fragments may collide, which causes information loss. For this reason collisions are considered unwanted and they are generally believed to decrease a performance. Our goal was, contrary to the general believe, test whether collisions can have positive impact on the performance. For this purpose we designed several similarity models based on fragments. In order to make testing and evaluation easy we implemented testing environ- ment. Results of our experiments prove that some collisions can outperform commonly used methods. Moreover some collisions in a specific model can lead to a performance of AUC over 0.99. 1
|
|
|
Detecting elementary particles with Timepix3 detector
Meduna, Lukáš ; Mráz, František (advisor) ; Hric, Jan (referee)
Detecting elementary particles and observing accompanying events in particle colliders is one of the most important field of current research in experimental physics. TimePix and its successor TimePix3 are types of the currently used detectors which are placed beside other in ATLAS experiment conducted by Eu- ropean Organization for Nuclear Research. Such detectors can produce huge amount of data about passing particles at high rate. The goal of the thesis is to develop methods for detecting and classification of elementary particles observed by detector network ATLAS-TPX3. Suitable methods for clustering and/or classification based on semi-labelled data should be identified or new one should be developed. The proposed methods will be implemented and their performance on real data will be evaluated. The results will also include an implementation of framework for preprocessing low level data from detector network ATLAS-TPX3 in real-time and creating outputs that are suitable for subsequent physics investigation (e.g. ROOT framework files) includ- ing the proposed or future methods for particle classification. 5
|
|
|
Echo state networks and their application in time series prediction
Savčinský, Richard ; Mráz, František (advisor) ; Matzner, Filip (referee)
Recurrent neural networks (RNN) enable to model dynamical sys- tems with variable input length. Their disadvantage is in inherently difficult trai- ning which means adjusting weights of connections between neurons connected in the network. Echo state networks (ESN) are a special type of RNN which are by contrast trainable rather simply. They include a reservoir of neurons whose state reflect the history of all signals in the network and that is why this type of network is suitable for simulation and prediction of time series. To maximize the computational power of ESN, very precise adjusting and experimenting are required. Because of that, we have created a tool for building and testing such networks. We have implemented a time series forecasting task for the purpose of examination of our tool. We have focused on stock price prediction, which repre- sents an uncertain and complicated area for achieving precise results in. We have compared our tool to other tools and it was comparably successful. 1
|
|
|
Echo state networks and their application in time series prediction
Savčinský, Richard ; Mráz, František (advisor) ; Matzner, Filip (referee)
Recurrent neural networks (RNN) enable to model dynamical sys- tems with variable input length. Their disadvantage is in inherently difficult trai- ning which means adjusting weights of connections between neurons connected in the network. Echo state networks (ESN) are a special type of RNN which are by contrast trainable rather simply. They include a reservoir of neurons whose state reflect the history of all signals in the network and that is why this type of network is suitable for simulation and prediction of time series. To maximize the computational power of ESN, very precise adjusting and experimenting are required. Because of that, we have created a tool for building and testing such networks. We have implemented a time series forecasting task for the purpose of examination of our tool. We have focused on stock price prediction, which repre- sents an uncertain and complicated area for achieving precise results in. We have compared our tool to other tools and it was comparably successful.
|
|
|
Klasifikace na množinách bodů v 3D
Střelský, Jakub ; Mráz, František (advisor) ; Šikudová, Elena (referee)
Increasing interest for classification of 3D geometrical data has led to discov- ery of PointNet, which is a neural network architecture capable of processing un- ordered point sets. This thesis explores several methods of utilizing conventional point features within PointNet and their impact on classification. Classification performance of the presented methods was experimentally evaluated and com- pared with a baseline PointNet model on four different datasets. The results of the experiments suggest that some of the considered features can improve clas- sification effectiveness of PointNet on difficult datasets with objects that are not aligned into canonical orientation. In particular, the well known spin image rep- resentations can be employed successfully and reliably within PointNet. Further- more, a feature-based alternative to spatial transformer, which is a sub-network of PointNet responsible for aligning misaligned objects into canonical orientation, have been introduced. Additional experiments demonstrate that the alternative might be competitive with spatial transformer on challenging datasets. 1
|
|
|
Comparison of signature-based and semantic similarity models
Kovalčík, Gregor ; Lokoč, Jakub (advisor) ; Mráz, František (referee)
Content-based image retrieval and similarity search has been investigated for several decades with many different approaches proposed. This thesis fo- cuses on a comparison of two orthogonal similarity models on two different im- age retrieval tasks. More specifically, traditional image representation models based on feature signatures are compared with models based on state-of-the-art deep convolutional neural networks. Query-by-example benchmarking and tar- get browsing tasks were selected for the comparison. In a thorough experimental evaluation, we confirm that models based on deep convolutional neural networks outperform the traditional models. However, in the target browsing scenario, we show that the traditional models could still represent an effective option. We have also implemented a feature signature extractor into the OpenCV library in order to make the source codes available for the image retrieval and computer vision community. 1
|
|
|
Using Cellular Automata for Data Compression
Polák, Marek ; Trunda, Otakar (advisor) ; Mráz, František (referee)
In this thesis we research the possibilities of using cellular automata for lossless data compression. We describe the classification of cellular automata and their current usage. We study the properties of various types of elementary cellular automata (i.e. Wolfram rules), describe their equivalence classes, the ways of forward as well as backward simulation, we examine the rules with interesting behavior. The states provided by these rules are evaluated in terms of their orderliness (e.g. the ratio of living cells or approximation of entropy). We implement some standard compression algorithms and compare them in terms of usability for best rated states. By application of acquired knowledge we propose a new compression algorithm, test it on text and image data and compare the results with traditional compression algorithms. Powered by TCPDF (www.tcpdf.org)
|
guest ::
