This work focuses on the design of a system that identifies frequently discussed product features in product reviews, summarizes them, and displays them to the user in terms of sentiment. The work deals with the issue of natural language processing, with a specific focus on Czech languague. The reader will be introduced the methods of preprocessing the text and their impact on the quality of the analysis results. The identification of the mainly discussed products features is carried out by cluster analysis using the K-Means algorithm, where we assume that sufficiently internally homogeneous clusters will represent the individual features of the products. A new area that will be explored in this work is the representation of documents using the Word embeddings technique, and its potential of using vector space as input for machine learning algorithms.

In the thesis, we explore the prospects of creating adversarial examples using various generative models. We design two algorithms to create unrestricted adversarial examples by perturbing the vectors of latent representation and exploiting the target classifier's decision boundary properties. The first algorithm uses linear interpolation combined with bisection to extract candidate samples near the decision boundary of the targeted classifier. The second algorithm applies the idea behind the FGSM algorithm on vectors of latent representation and uses additional information from gradients to obtain better candidate samples. In an empirical study on MNIST, SVHN and CIFAR10 datasets, we show that the candidate samples contain adversarial examples, samples that look like some class to humans but are classified as a different class by machines. Additionally, we show that standard defence techniques are vulnerable to our attacks.

The aim of this thesis is to minimize manual work needed to create training data for text classification tasks. Various research areas including weak supervision, interactive learning and transfer learning explore how to minimize training data creation effort. We combine ideas from available literature in order to design a comprehensive text classification framework that employs keyword-based labeling instead of traditional text annotation. Keyword-based labeling aims to label texts based on keywords contained in the texts that are highly correlated with individual classification labels. As noted repeatedly in previous work, coming up with many new keywords is challenging for humans. To accommodate for this issue, we propose an interactive keyword labeler featuring the use of word similarity for guiding a user in keyword labeling. To verify the effectiveness of our novel approach, we implement a minimum viable prototype of the designed framework and use it to perform a user study on a restaurant review multi-label classification problem.

A B S T R A C T ( E N G L I S H ) Classification and cataloguing of folk narrative has been one of the important topics of folklore comparative studies. The present thesis examines the topic from the theoretical and methodological point of view in its first part and from the appliqued point of view in its second part. The introduction of the theoretical part describes the genre of folk legend on the basis of textual (content, form, structure) and contextual (belief, function, distribution) criteria. Next chapter pursues an analysis of classificatory systems and catalogues of folk narrative of the whole world. The chapter is divided into two main groups according to typological or structural principles. We focus on the development of international folktale catalogue of Aarne-Thompson(-Uther) and its influence, especially when encountered with catalogues of Non-European narratives. We also deal with Czech and Slovak works in cataloguing and our main concern is cataloguing of European folk legends. The content of the following chapter is an overview and evaluation of sources selected as a material for our catalogue. The practical part of the work is a proposal and processing of typological catalogue of Czech belief legends, which is intended to serve as a tool for folklorists and scholars of relative...

Mgr. Jan Froněk, disertační práce, Katedra andragogiky a personálního řízení FF UK v Praze Abstract The dissertation thesis presents varieties of games and simulations in the field of adult education and specializes particularly in role-play and simulation classifications, with the emphasis on presenting case examples of the applications of the methods. To reach the objectives of the dissertation, first of all resources analysis had been made (both scientific resources and practical toolkits). Then the acquired knowledge was reframed into a terminological framework that had existed before, yet had not been defined specifically enough and explicitly for the field of adult education. The second pillar of the thesis represents a survey that was conducted among adult educators. The thesis consists of six main chapters: it is introduced by a general chapter on games, plays and their characteristics and definition (also from adult educator's perspective). A chapter on simulation and gaming classification follows (Chap. 2), from which the largest part of this work is drawn: a detailed categorization of role-play (Chap. 3) and simulation (Chap. 4) varieties. Some of the Czech as well as international simulation and gaming organisations are mentioned (Chap. 5). In the 6th chapter, results from a survey are...

Matzke, M. Comparison of Computational Complexity of Selected Data Mining Algorithms, Diploma Thesis. Brno, 2018. This diploma thesis deals with the comparison of the time complexity and the success of the classification of selected algorithms for mining knowledge from data with focus on neural networks and optimal settings for work execution. In the theoretical part, it is essential to get acquainted with the distribution of algorithms, their functionality and complexity. Then follows the selection of algorithms with focus on neural networks and their settings, especially hidden layers, momentum and learning rate. Another part deals with data used for experimental testing, which are both nominal and numerical data, and also real or generated. Also included is the accuracy of measurement and performance measurement of the two assemblies used to test individual experiments. The third part is the testing of the time complexity and the percentage success of the algorithms and the output especially in graphical form followed by analysis and recommendations from the results with focus on the optimal setting against the automatic and initial settings.

Work deals with code detection from control image. The document presents relevant image processing techniques dealing with a noise reduction, thresholding, color models, object segmentation and OCR. This project examines advantages and disadvantages of two selected methods for object segmentation and introduces developed system for object segmentation. The developed system for object segmentation and classification is realized, evaluated and results are discussed in details.

An AdaBoost algorithm for construction of strong classifier from several weak hypotesis will be presented in this work. Theoretical background of the algorithm and the method of construction of strong classifiers will be explained. WaldBoost extension to the algorithm will be described. The thesis deals with image features that are often used as element of weak classifiers. Brief introduction to pattern recognition in context of computer vision will be outlined in the begining of the work. Also some widely used methods of classifier training will be presented. An object detection library based on AdaBoost classifiers was developed as part of the work. The library was used in implementation of software that in praktice demonstrates object detection in videosquences. Last part of the work describes tool for training of AdaBoost classifiers.

This thesis examines adversarial examples in machine learning, specifically in the im- age classification domain. State-of-the-art deep learning models are able to recognize patterns better than humans. However, we can significantly reduce the model's accu- racy by adding imperceptible, yet intentionally harmful noise. This work investigates various methods of creating adversarial images as well as techniques that aim to defend deep learning models against these malicious inputs. We choose one of the contemporary defenses and design an attack that utilizes evolutionary algorithms to deceive it. Our experiments show an interesting difference between adversarial images created by evolu- tion and images created with the knowledge of gradients. Last but not least, we test the transferability of our created samples between various deep learning models. 1

The Pilsen region suffers the highest incidence of kidney tumours worldwide. Approximately 240 new cases diagnosed as C64 (malignant renal tumours outside the pelvis) were recorded in this region of about 580,000 inhabitants in 2015. Clear renal cell carcinoma has long held first place as the most common tumour, with papillary renal cell carcinoma (pRCC) being the second most frequently operated kidney tumour at the Urology Department of the University Hospital in Pilsen. The 2016 WHO classification of kidney tumours recognizes officially only the stratification of pRCC to type 1 (pRCC1) and type 2 (pRCC2). Unfortunately, the current division does not correspond with knowledge derived from everyday practice. Most clinical trials involving pRCC do not differentiate between the subtypes, adhering only to the official type 1 and 2 divisions and the atypical papillary forms being excluded from their studies. We therefore have to face the question of whether the histological pRCC subtype affects the risk of recurrence, or death, in surgically treated patients. The aim of this dissertation work is to take into consideration also all other papillary types which differ from characterization of pRCC1 and pRCC2. The analyses of a group of patients with surgically treated and histologically verified pRCC at...