|
|
Optimization of control using reinforcement learning on the Robocode platform
Pastušek, Václav ; Myška, Vojtěch (referee) ; Burget, Radim (advisor)
This master's thesis focuses on optimizing the control of a tank robot in the Robocode environment using reinforcement learning. The complexity of this problem falls into the EXPSPACE class, presenting a challenge that cannot be underestimated. The theoretical part of the thesis meticulously examines the Robocode platform, concepts of reinforcement learning, and relevant algorithms, while the practical part focuses on optimizing the agent, implementing reinforcement learning algorithms, and creating a user-friendly interface for easy training and testing of models. A total of 64 models were trained and tested as part of the thesis, with their data and parameters compared and presented in accompanying databases and graphs. The best results in terms of average hits per episode were achieved by models labeled v0.8.0 and v1.0.0. The first model exhibited a certain ability to evade shots, while the second model showed more successful hits.
|
|
|
Automated segmentation of the diadochokinetic task for remote monitoring of hypokinetic dysarthria
Svojanovský, Jan ; Mekyska, Jiří (referee) ; Kováč, Daniel (advisor)
The study describes health problems associated with Parkinson’s disease, especially hypokinetic dysarthria. It also points out the subjective and objective methods used to determine the severity of the disease. One of these methods is a diadochokinetic (DDK) task based on rapid syllable repetition to test the functionality of the articulatory apparatus (e.g., tongue, lips, or vocal cords). Correct speech production can also be examined by a speech therapist in the 3F test, which scores the severity of disorders in different areas of speech production. Next, the approaches of other authors, also dealing with the automated search of syllables in the speech signal, are described. The thesis also discusses some features of human speech that are needed for training a machine learning model. These features were computed for each of the 30 ms segments of a DDK task. The main goal is the automated detection and classification of [Pa]-[Ta]-[Ka] syllables in the recordings. For this purpose, an algorithm using a logistic regression was applied. The resulting average accuracy of syllable detection in the recordings was 89.4 %, average sensitivity 59.0 % and average specificity 93.79 %. The identification of individual syllable types was successful with an average accuracy of 90.78 %, an average sensitivity of 59.0 % and an average specificity of 95.39 %. Considering that the predicted onset was not located directly on the manually annotated onset, but in its close vicinity (up to ±3 segments), the average detection sensitivity and average syllable type classification sensitivity were 96.9 % and 85.1 % respectively, with an average difference between manually annotated and automatically segmented syllable onsets of 10.35 ms. The average accuracy of classification of speakers into healthy and PN patients using logistic regression (with speech parameters obtained after automated segmentation) was only 43.92 %, sensitivity 70.0 % and specificity 30.61 % (threshold 70 %). Using linear regression, the clinical scores of the 3F test were predicted. For faciokinesis, the root mean square error (RMSE) was 2.764 after manual syllable annotation and 3.271 after automated segmentation. The RMSE values for phonetics were 3.657 (manual) and 0.753 (automated). The developed algorithm can detect syllables in DDK tasks with relative success, and thus it is possible to determine parameters quantifying speech disorders with low differences with manual segmentation. If the recordings of DDK tasks meet the conditions for computing all these parameters, the algorithm could be used to classify speakers into healthy subjects and PN patients, for whom it could additionally assess the severity of dysarthria.
|
|
|
Support tool for initial phase of penetration testing
Žáček, Dominik ; Gerlich, Tomáš (referee) ; Sikora, Pavel (advisor)
This thesis deals with the development of an advanced tool designed to make team penetration testing more efficient. The tool works by automatically assigning tasks to penetration testers based on skills and historical performance. The theoretical part of the thesis analyzes in detail various methods for solving the assignment problem, in particular the Hungarian method and linear programming. The theoretical part continues with the design of a two-step algorithm for task assignment. Then, the principle of the neural networks underlying the second step of the assignment is described in detail. Unique methods for generating two datasets have also been developed as part of the work. An interface for task assignment has been implemented and metrics to determine the quality of the assignment have been proposed. The result is a tool that significantly streamlines the assignment of tasks to penetration testers and increases the overall efficiency of penetration testing teams.
|
|
|
Design of signals for nonlinear system modeling
Kuba, Michael ; Ištvánek, Matěj (referee) ; Miklánek, Štěpán (advisor)
This Master’s thesis is focused on training signals for nonlinear system modeling using deep learning. A theoretical introduction to the problem is described, including an initial description of signals and nonlinear distortion audio effects. The goal of the thesis is to design a set of artificial training signals for creating models of distortion guitar effects or tube guitar amplifiers. The designed set of artificial training signals is then processed, utilizing guitar and bass guitar effects, and using a recurrent neural network their models are trained. The quality of the resulting models is afterwards compared with the quality of the models trained with the help of a reference training signal, composed of electric guitar recordings, and signals from commercially available devices. The comparison is carried out in accordance to objective metrics and with subjective evaluation by the MUSHRA listening test.
|
|
|
Current Trends in Data Analytics and their Successful Enterprise Applications
RUBÁŠOVÁ, Anna
This bachelor thesis focuses on current trends in data analytics and their application to business practice. The thesis describes the development of data analytics and machine learning methods that are used to mine knowledge from data. A specific example demonstrating the use of machine learning deals with predicting the outcome of a binary classification problem, where historical data is used to predict future trends. The selected methods are implemented using a "no-code" web-based tool. A comparative analysis of the implemented methods is performed in the thesis. The methods are evaluated on the basis of their accuracy and computational efficiency, and then the most appropriate one is identified. Finally, the selected method is further optimized using a threshold value to achieve the best results.
|
|
|
Encrypted video-stream identification
MACÁK, Tomáš
The aim of this thesis is to create a data set of measured encrypted video streams and subsequently try to discover if it is possible to identify the content of those streams. In the theoretical part the on - demand video streaming is introduced and then suitable machine learning models applicable to solve this problem are presented. The works focused on a similar topic are presented next. In followed practical part the already mentioned data set is created. This set is then analysed and it is determined if there is a way how to represent those measured video streams for later content identification with use of statistical and machine learning models. In the last part of this chapter the machine learning models for classification and similarity detection are implemented and trained. The models are then tested and the results are summarised and compared.
|
|
|
Anomaly and threat detection in audit logs using machine learning
Ludes, Adam ; Ježek, Štěpán (referee) ; Tomašov, Adrián (advisor)
Tato práce představuje softwarové architektury založené na cloudu, techniky detekce anomálií, strojové učení a analýzu dat za účelem vytvoření modelu pro detekci anomálií v audit lozích z Red Hat OpenShift Container Platform. Jsou představeny statistické metody a analýza časových řad pro detekci anomálií, zatímco jsou implementovány a hodnoceny modely strojového učení a techniky předzpracování dat. Výsledky ukazují omezení tradičních modelů při zpracování anomálií v hluboce vnořených datech, zatímco model zpracovávající přirozený jazyk prokazuje robustní výkon. Tato práce poskytuje cenné poznatky a může být použita jako reference pro výzkum i praxi v oblasti softwarových architektur založených na cloudu, detekce anomálií, strojového učení a analýzy dat.
|
|
|
DEEP LEARNING FOR SINGLE-VOXEL AND MULTIDIMENSIONAL MR-SPECTROSCOPIC SIGNAL QUANTIFICATION, AND ITS COMPARISON WITH NONLINEAR LEAST-SQUARES FITTING
Shamaei, Amirmohammad ; Latta,, Peter (referee) ; Kozubek, Michal (referee) ; Jiřík, Radovan (advisor)
Pro získání koncentrace metabolitů ve vyšetřované tkáni ze signálů magnetické rezonanční spektroskopie (MRS) je nezbytné provézt předzpracování, analýzu a kvantifikaci MRS signálu. Rychlý, přesný a účinný proces zpracování (předzpracování, analýza a kvantifikace) MRS dat je však náročný. Tato práce představuje nové přístupy pro předzpracování, analýzu a kvantifikaci MRS dat založené na hlubokém učení (DL). Navržené metody potvrdily schopnost použití DL pro robustní předzpracování dat, rychlou a efektivní kvantifikaci MR spekter, odhad koncentrací metabolitů in vivo a odhad nejistoty kvantifikace. Navržené přístupy výrazně zlepšily rychlost předzpracování a kvantifikace MRS signálu a prokázaly možnost použití DL bez učitele. Z hlediska přesnosti byly získány výsledky srovnatelné s tradičními metodami. Dále byl zaveden standardní formát dat, který usnadňuje sdílení dat mezi výzkumnými skupinami pro aplikace umělé inteligence. Výsledky této studie naznačují, že navrhované přístupy založené na DL mají potenciál zlepšit přesnost a efektivitu zpracování MRS dat pro lékařskou diagnostiku. Disertační práce je rozdělena do čtyř částí: úvodu, přehledu současného stavu výzkumu, shrnutí cílů a úkolů a souboru publikací, které představují autorův přínos v oblasti aplikací DL v MRS.
|
|
|
Automatic diagnosis of the 12-lead ECG using deep learning
Blaude, Ondřej ; Smital, Lukáš (referee) ; Provazník, Valentine (advisor)
The aim of this diploma thesis is to investigate the problematics of automatic ECG diagnostics, namely on twelve-lead recordings. In the first chapter the heart and its electrical activity measurement is described shortly. In addition to that, the abnormalities which are going to be classified in this thesis are also briefly described. In the second chapter, it is described how the ECG was diagnosed earlier, by classical methods that preceded deep learning. Some of the shortcomings that the classical methods have compared to deep learning are also described here. The third part already pays attention to deep learning itself, and its contribution and advantages compared to classical methods. Convolutional neural networks and their individual blocks are also described here, later attention is paid to selected architectures that were used in some studies. The fourth chapter already focuses on the practical part, in which the data used from the PhysioNet database, the proposed algorithm and its implementation are described in more detail. In the fifth chapter the results are discussed and compared to the corresponding publications.
|
|
|
Correction of the concept of drift in prediction models
Michálková, Eva ; Provazník, Valentine (referee) ; Schwarzerová, Jana (advisor)
The main goal of this bachelor thesis is the analysis of concept drift in metabolomics. Concept drift is an undesirable phenomenon and can be caused by nonstationary data. It can have a negative impact on the performance and reliability of predictive modelling. This challenge can be solved by concept drift detection and subsequent correction. One of the fields where this issue has recently emerged is metabolomic diagnostics. Metabolomic data analysis can lead to early detection of several serious diseases, which can help with the recovery process. When diagnosing an illnes predictive models present a way to make the process more efficient, faster and give the option of personalization. The first part of this thesis specifies concept drift, it’s detection and correction methods and the importance of metabolomics and prediction models. The second part deals with the implementation of some available algorithms for concept drift detection and correction and the implementation of automatic concept drift correction. Finally, in the second part results and their discussion are described.
|
guest ::
