|
|
Deep Book Recommendation
Gráca, Martin ; Kolář, Martin (referee) ; Hradiš, Michal (advisor)
This thesis deals with the field of recommendation systems using deep neural networks and their use in book recommendation. There are the main traditional recommender systems analysed and their representations are summarized, as well as systems with more advanced techniques based on machine learning. The core of the thesis is to use convolutional neural networks for natural language processing and create a hybrid book recommendation system. Suggested system includes matrix factorization and make recommendation based on user ratings and book metadata, including texts descriptions. I designed two models, one with bag-of-words technique and one with convolutional neural network. Both of them defeat baseline methods. On the created data set, that was created from the Goodreads, model with CNN beats model with BOW.
|
|
|
Low-Dimensional Matrix Factorization in End-To-End Speech Recognition Systems
Gajdár, Matúš ; Grézl, František (referee) ; Karafiát, Martin (advisor)
The project covers automatic speech recognition with neural network training using low-dimensional matrix factorization. We are describing time delay neural networks with factorization (TDNN-F) and without it (TDNN) in Pytorch language. We are comparing the implementation between Pytorch and Kaldi toolkit, where we achieve similar results during experiments with various network architectures. The last chapter describes the impact of a low-dimensional matrix factorization on End-to-End speech recognition systems and also a modification of the system with TDNN(-F) networks. Using specific network settings, we were able to achieve better results with systems using factorization. Additionally, we reduced the complexity of training by decreasing network parameters with the use of TDNN(-F) networks.
|
|
|
Document Quality Enhancement
Trčka, Jan ; Zemčík, Pavel (referee) ; Juránek, Roman (advisor)
The aim of this work is to increase the accuracy of the transcription of text documents. This work is mainly focused on texts printed on degraded materials such as newspapers or old books. To solve this problem, the current method and problems associated with text recognition are analyzed. Based on the acquired knowledge, the implemented method based on GAN network architecture is chosen. Experiments are a performer on these networks in order to find their appropriate size and their learning parameters. Subsequently, testing is performed to compare different learning methods and compare their results. Both training and testing is a performer on an artificial data set. Using implemented trained networks increases the transcription accuracy from 65.61 % for the raw damaged text lines to 93.23 % for lines processed by this network.
|
|
|
Removing noise in images using deep learning methods
Strejček, Jakub ; Jakubíček, Roman (referee) ; Vičar, Tomáš (advisor)
This thesis focuses on comparing methods of denoising by deep learning and their implementation. In the last few years, it has become clear that it is not necessary to have paired data, as for noisy and clean pictures, to train convolution neural networks but it is sufficient to have only noisy pictures for denoising in particular cases. By using methods described in this thesis it is possible to effectively remove i.e. additive Gaussian noise and what more, it is possible to achieve better results than by using statistic methods, which are being used for denoising these days.
|
|
|
The influence of first CNN layer initialization on training convergence
Krejsa, Jiří ; Věchet, Stanislav ; Chen, K.S.
During evaluation of convolution neural networks on the task of sign language single hand alphabet classification we have discovered that in small but not negligible number of cases the training of the network does not converge at all. This paper investigates the problem that we believe is independent of the application. While the true cause of training divergence was not discovered, we can offer the reader an easy solution from practical point of view – initialization of the first CNN layer using pretrained networks parameters.
|
|
|
Exploitation of Neural Networks for Fusion of Image and Non-Image Data
Pařilová, Michaela ; Reich, Bořek (referee) ; Zemčík, Pavel (advisor)
This bachelor's thesis deals with exploitation of neural networks for fusion of image and non-image data. It primarily focuses on comparing detection algorithms for image data (data obtained from a camera) with detection algorithms using the fusion of image data and non-image data (data obtained from a camera and radar). As part of the comparison of these two approaches, an own fusion model was created, which was subsequently tested on the selected data set.
|
|
|
Design and implementation of the robotic platform for an experimental laboratory task
Juříček, Martin ; Matoušek, Radomil (referee) ; Parák, Roman (advisor)
Pokročilá robotika se nemusí vždy pouze pojit s Průmyslem 4.0, nýbrž nachází své uplatnění i kupříkladu v konceptu Smart Hospital. Pokrok v této oblasti umocnilo onemocnění koronaviru (COVID-19), přičemž každé ulehčení práce zdravotnímu personálu je vítáno. V rámci této diplomové práce byla navrhnuta a implementována experimentální robotická platforma s hlavní funkcí stěru vzorků z předsíně dutiny nosní. Robotická platforma představuje kompletní integraci softwaru a hardwaru, kde má operátor přístup k webově založené aplikaci a může ovládat řadu funkcí. Opomenout nelze také zvýšenou bezpečnost a kolaborativní přístup. Výsledkem práce je funkční prototyp robotické platformy, který je možné dále rozšiřovat například v podobě použití alternativních technologií, rozšíření bezpečnosti či klinického testování a studie.
|
|
|
Suppression of the responsive component of electrodermal activity
Vraný, Jakub ; Vičar, Tomáš (referee) ; Kolářová, Jana (advisor)
Electrodermal acitivity is a kind of electrochemical signal generated with relation to activity of the autonomic nervous system that stimulates the sweat glands. In this way, is it possible to measure the activity of the sympathetic part of the nerve systém and evaluate the cognitive stress of the treated person, which is manifested by responsive signals in EDA record, respectively to increased occurence of responses. The aim of this work is to design a deep learning algorithm for the identification of this component in the record of data taken from UBMI database. The recordings contain a sequence of measurements the conductance of the skin of patient, who was subjected alternately to the states of rest and subsequently a state of mental stress. The data were annotated according to presence of the responsive components occuring in the records of EDA. Subsequently, a suitable deep learning algorithm was implemented in order to classify the responsive components in the measured EDA signal. The neural network model has been taught, optimized and implemented on the measurement samples using annotated data. The obtained results data were statistically evaluated to qualify the success of the classification of responsive components and differences in the records of mental calm and stress. The results of the classification and comparison of EDA records measured at different conditions of the patient were discussed subsequently.
|
|
|
Larval morphology, phylogeography and automatic identification of chosen flower chafer beetles (Scarabaeidae: Cetoniinae)
Vondráček, Dominik ; Šípek, Petr (advisor) ; Bezděk, Aleš (referee) ; Kundrata, Robin (referee)
Currently, over 4.300 species of flower chafer beetles (Scarabaeidae: Cetoniinae) are described in more than 485 genera, with the number of genera, species and subspecies increasing by dozens of new taxa each year. Especially in the past, some of the species descriptions were relatively vague and short operating only with the coloration of beetles, or with extremely subtle differences on the male genitalia without any support of other data and analyses. In this dissertation, I focused on the use of various data and methodological approaches that can help understand the evolutionary processes within this group and its complicated taxonomy and systematics, which is also still very unstable even at higher taxonomic ranks. In two works we studied the morphology of immature stages of flower chafers and their bionomy. In the case of the Taenioderini tribe, whose immature stages were not known until then, we found surprisingly significant morphological variability in the eight described species, which is unusual in larval stages of flower chafers. In the second work, we focused on the genus Oxythyrea. We described the larvae of nine of the ten currently known species and confronted the obtained data with the already existing descriptions of the larvae of the Leucocelina subtribe, to which the studied...
|
|
|
Automated Human Recognition From Image Data
Dobiš, Lukáš
This paper describes an approach for automated human recognition by using convolutional neural networks (CNN) to perform facial analysis of persons face from image data. The predicted biometric indicators are following: age, gender, facial landmarks and facial expression. Network architectures with pretrained weights for each task are described. Script of interconnected CNN is explained and its results support further proposed expansion plans for live video inference.
|
guest ::
