|
|
Classification of board defects in semiconductor manufacturing
Jašek, Filip ; Vágner, Martin (referee) ; Dřínovský, Jiří (advisor)
This diploma thesis focuses on detecting defects in semiconductor wafer manufacturing. It explores methods for identifying faulty chips and controlling yield during production. To classify defects machine learning techniques are used. Initially, ResNet18 architecture was used for inference, but low accuracy was attributed to limited input data. Transfer learning with ResNet50v2 was then attempted, resulting in improved metric with different dataset. Hyperparameter tuning and data augmentations were also explored. The study found that autoencoders for data compression during inference increased speed but led to degraded evaluation metrics.
|
|
|
Machine Learning from Intrusion Detection Systems
Dostál, Michal ; Očenášek, Pavel (referee) ; Hranický, Radek (advisor)
The current state of intrusion detection tools is insufficient because they often operate based on static rules and fail to leverage the potential of artificial intelligence. The aim of this work is to enhance the open-source tool Snort with the capability to detect malicious network traffic using machine learning. To achieve a robust classifier, useful features of network traffic were choosed, extracted from the output data of the Snort application. Subsequently, these traffic features were enriched and labeled with corresponding events. Experiments demonstrate excellent results not only in classification accuracy on test data but also in processing speed. The proposed approach and the conducted experiments indicate that this new method could exhibit promising performance even when dealing with real-world data.
|
|
|
Interpreting the learning process of an atrial fibrillation classifier
Lichtblauová, Anna ; Ředina, Richard (referee) ; Novotná, Petra (advisor)
In the theoretical part of the bachelor thesis the problems of atrial fibrillation (AF) detection and principles of convolutional neural networks (CNN) are discussed. Next, two classifiers were created in the practical part. The first was designed to classify sinus rhythm, atrial fibrillation and other pathologies, while the second further distinguished the category "atrial fibrillation" according to whether it was present in the whole recording or only in a part of it. The resulting accuracies are 82.12 \% and 85.14 \% for the first and second classifiers, respectively.
|
|
|
Identification of specified segments in the audio signal using machine learning
Pařízek, Radim ; Galáž, Zoltán (referee) ; Zvončák, Vojtěch (advisor)
The bachelor thesis deals with the design of a system for the identification of natural environmental sounds in audio recordings. The datasets and models used for this type of tasks are surveyed and their structure is described. A system for the identification of sounds in one layer and in two layers has been proposed for seven selected labels. The classifier used for this system was created by fine-tuning a transformer model from the Hugging Face platform. The results of two training approaches and one identification system were evaluated.
|
|
|
Approximation of functions determining colony activity using neural networks
Nevláčil, Jakub ; Ligocki, Adam (referee) ; Honec, Peter (advisor)
Bees as a primary pollinator are an indispensable contribution to global agriculture and food production. However, their numbers have been constantly declining in recent times, primarily due to climate change, parasites or the effect of pesticide use. Understanding their behavior and reliably determine their activity and health could significantly prevent or slow down their decline. That is why this work deals with the development of a device for the acquisition of useful data from beehives, which could be used to determine the activity and health of the bees. Furthermore, this work deals with analysis of the accumulated data using machine learning methods with an emphasis on determining the activity and health of the bees.
|
|
|
Playing Games Using Neural Networks
Buchal, Petr ; Kolář, Martin (referee) ; Hradiš, Michal (advisor)
The aim of this bachelor thesis is to teach a neural network solving classic control theory problems and playing the turn-based game 2048 and several Atari games. It is about the process of the reinforcement learning. I used the Deep Q-learning reinforcement learning algorithm which uses a neural networks. In order to improve a learning efficiency, I enriched the algorithm with several improvements. The enhancements include the addition of a target network, DDQN, dueling neural network architecture and priority experience replay memory. The experiments with classic control theory problems found out that the learning efficiency is most increased by adding a target network. In the game environments, the Deep Q-learning has achieved several times better results than a random player. The results and their analysis can be used for an insight to reinforcement learning algorithms using neural networks and to improve the used techniques.
|
|
|
Assessment of Parkinson’s Disease Based on Acoustic Analysis of Hypokinetic Dysarthria
Galáž, Zoltán ; Brezany, Peter (referee) ; Sklenář, Jaroslav (referee) ; Mekyska, Jiří (advisor)
Hypokinetická dysartrie (HD) je častým symptomem vyskytujícím se až u 90% pacientů trpících idiopatickou Parkinsonovou nemocí (PN), která výrazně přispívá k nepřirozenosti a nesrozumitelnosti řeči těchto pacientů. Hlavním cílem této disertační práce je prozkoumat možnosti použití kvantitativní paraklinické analýzy HD, s použitím parametrizace řeči, statistického zpracování a strojového učení, za účelem diagnózy a objektivního hodnocení PN. Tato práce dokazuje, že počítačová akustická analýza je schopná dostatečně popsat HD, speciálně tzv. dysprozodii, která se projevuje nedokonalou intonací a nepřirozeným tempem řeči. Navíc také dokazuje, že použití klinicky interpretovatelných akustických parametrů kvantifikujících různé aspekty HD, jako jsou fonace, artikulace a prozodie, může být použito k objektivnímu posouzení závažnosti motorických a nemotorických symptomů vyskytujících se u pacientů s PN. Dále tato práce prezentuje výzkum společných patofyziologických mechanizmů stojících za HD a zárazy v chůzi při PN. Nakonec tato práce dokazuje, že akustická analýza HD může být použita pro odhad progrese zárazů v chůzi v horizontu dvou let.
|
|
|
Automatic Cryptocurrencies Trading
Vorobiev, Nikolaj ; Hrubý, Martin (referee) ; Rozman, Jaroslav (advisor)
This thesis focuses on the trading in the cryptocurrency market. The theoretical part of the thesis describes the principles of trading, technical analysis, trading systems and recurrent neural networks. After conducting a search of brokers, Binance is chosen as a trading broker and real-time data provider; CryptoDataDownload is chosen as a historical data provider. After getting acquainted with the technologies used, elements of information trading systems are designed, enabling communication with remote servers and with each other, for the purpose of trading, obtaining and concurrent processing of user's, historical or real-time data. The resulting systems should provide to the user manual, semi-automatic (according to the plan) or automatic (according to the decisions of recurrent neural network, learned on historical data) trading and ability to respond to a change in the market. Furthermore, the thesis moves to the practical level, including implementation and experiments on created systems. In the final part of the thesis, the results are evaluated and the possibilities for improvement and expansion are described.
|
|
|
Automatic detection of tool fracture in metal sheet punching
Kluz, Jan ; Rajchl, Matej (referee) ; Brablc, Martin (advisor)
This Bachelor thesis deals with the design and subsequent implementation of the realtime fault detection system during the sheet metal punching process with a tool of small dimensions (0.5 × 12 mm). The proposed system is important for significant ease of the operator's work, acceleration of the process of production, as well as saving of the company finance budget. The first part of this thesis deals with the theoretical background of the studied issue. The following part is a brief theoretical introduction to the field of digital signal processing. The next chapter presents methods developed for fault signals detection including speed enhancing and data flow reducing algorithms. The main examined methods were: frequency peaks, frequency bands, autocorrelation, frequency correlation methods and machine learning including deep machine learning. Deep machine learning of the neural network achieved the best results overall. Features from time and frequency domain were used for purposes of creating the classification model using machine learning. The possibility of developing the predictive maintenance system is also described, including research of this area in a modern industry. Subsequently, the achieved results and their evaluation are presented. The end of this thesis is dedicated to the description of the implementation of classification system into realtime form and connecting this system to the punching press computer using Arduino Uno microcontroller and basic signal control electronics. The proposed system has been successfully assembled, tested and put into on-site testing.
|
|
|
Multi-Label Classification of Text Documents
Průša, Petr ; Očenášek, Pavel (referee) ; Bartík, Vladimír (advisor)
The master's thesis deals with automatic classifi cation of text document. It explains basic terms and problems of text mining. The thesis explains term clustering and shows some basic clustering algoritms. The thesis also shows some methods of classi fication and deals with matrix regression closely. Application using matrix regression for classifi cation was designed and developed. Experiments were focused on normalization and thresholding.
|
guest ::
