Ing. Zdeněk Buk, Ph.D.

Ing. Zdeněk Buk, Ph.D.

Ing. Miroslav Čepek, Ph.D.

Department of Applied Mathematics

The bachelor thesis deals with the estimation of the age of the human face captured in a three- dimensional facial scan. The estimation is performed using machine learning methods especially artificial neural networks. The thesis assesses the error rate of age estimation using different arti- ficial neural networks. It considers the different representations of the data that can be converted from the three-dimensional scan and analyzed, and their impact on the resulting age prediction. It also examines the impact of classification in relation to the gender of the person captured in the scan. Finally, it assesses the accuracy of the recognition of the age groups 15, 18 and 40 years.

Thesis on DSpace

Ing. Zdeněk Buk, Ph.D.

doc. Ing. Pavel Kordík, Ph.D.

Department of Theoretical Computer Science

This thesis deals with the topic of using neural networks in place of human player in playing computer game Asteroids. It describes how to use different types of artificial intelligence in this game. A clone of the game and fully recurrent neural network playing the game from human perspective are made in this work. Differential evolution, a subset of evolutionary algorithms, was used for learning purposes of the neural network. Furthermore this thesis experiments with different settings of the evolution and neural network. Results of said experimentation are compared with human behavior.

Thesis on DSpace

Ing. Zdeněk Buk, Ph.D.

Ing. Michal Štepanovský, Ph.D.

Department of Theoretical Computer Science

The problem of searching in uncategorized documents is that users are often presented with results that contain searched keywords, but are not relevant to the user. The goal of this work is to extend the documents with tags based on their content. To accomplish this several approaches were considered: Elasticsearch, Semaphore, LSTM, BERT. The objective of this thesis is to determine which technology has the most potential and provides the best results. All listed approaches were tested and evaluated. It was found that BERT models performed the best and satisfied all of the initial business requirements. Some improvements in the quality of classification with BERT were achieved by utilizing the initial model and manually classifying a small set of documents with a low confidence score.

Thesis on DSpace

Ing. Zdeněk Buk, Ph.D.

Ing. Jan Drchal, Ph.D.

Department of Theoretical Computer Science

This bachelor thesis deals with creating artificial inteligence for poker using artifical neural networks. Unlike other projects, that focus on artificial inteligence for games, neural network is not just one of many parts in big algorithm, but it decides which moves to take on its own. Players use reinforcement learning to improve their play. Tests show that neural network trained this way plays on similar level as recreation player.

Thesis on DSpace

Ing. Zdeněk Buk, Ph.D.

Ing. Michal Štepanovský, Ph.D.

Department of Computer Systems

The aim of this work is to explore the possibilites of remote control for the GoPro cameras, to design and implement a package of functions using the Wolfram Language, which will enable this functionality in the environment of Wolfram Mathmeatica. The resulting package offers a setting of the camera mode (for photographs and videos), a setting of the image parameters (resolution, exposure compensation, FPS), remote control of the camera and accessibility of the data.

Thesis on DSpace

Ing. Zdeněk Buk, Ph.D.

doc. RNDr. Pavel Surynek, Ph.D.

Department of Applied Mathematics

This thesis describes the design and implementation of a system for comparing the similarity of columns in an arbitrary database. We have shown that our system, based on recurrent neural networks, outperforms the industry standard TF-IDF method by \textbf{14.5\%}. We therefore, conclude that our system is capable of learning to effectively recognize the domain properties of data in the database. We deployed the described system in Atacama One, where it is responsible for Business Terms recommendations.

Thesis on DSpace

Ing. Zdeněk Buk, Ph.D.

Ing. Jan Drchal, Ph.D.

Department of Theoretical Computer Science

This thesis focuses on generating music using recurrent neural networks. Models with GRU and LSTM nodes in various configurations were trained on data with classical piano music stored in a MIDI format. Based on these data the model has generated new compositions. Firstly the ability to learn a specific composition and to generate identical copies was tested. Secondly experiments with generating new music pieces took place. Neural nets were able to generate copies that were identical to the original by 83 %. The result of this work is an analysis of several projects dealing with music generaction and creating a program capable of generating new compositions based on existing music pieces.

Thesis on DSpace

Ing. Zdeněk Buk, Ph.D.

Ing. Michal Štepanovský, Ph.D.

Department of Applied Mathematics

I present DeepRCar, a simplified self-driving radio controlled car platform that is controlled by deep neural networks. This car takes images from a front facing camera as its only input and produces steering commands as output. This thesis describes the entire process of its creation from hardware requirements, through the design of the control system, up to the selection and training of a convolutional neural network that manages its driving decisions. The network was trained in an end-to-end manner and learned to recognize useful road features, such as lane markings, when only camera images and corresponding steering angles were presented during training. The final system is capable of running at 20 frames per second on a Raspberry Pi 3.

Thesis on DSpace

Ing. Zdeněk Buk, Ph.D.

Ing. Michal Štepanovský, Ph.D.

Department of Applied Mathematics

This work focuses on the design and implementation of an environment suitable for training a neural neural network how to control a robot similar to a car. A neural network can be connected to the environment via TCP protocol and, thus, does not limit the implementation of neural network. The second part of this work is a script that teaches the neural network how to park a car using data from the car's cameras. This part serves as a evidence of the usability of the created environment for the purpose of teaching the neural network how to control a car-like robot. The final neural network consists of two parts, where the convolutional network localizes the parking spot from an image from a camera and output coordinates are fed to the recurrent neural network giving commands to the car. The environment is implemented in the game engine Unity 3D and script in Wolfram Mathematica.

Thesis on DSpace

Ing. Zdeněk Buk, Ph.D.

doc. RNDr. Ing. Marcel Jiřina, Ph.D.

Department of Theoretical Computer Science

The aim of this diploma thesis is to model customer behavior using suitable algorithms. I use models for processing time events and transaction data: Markov chains, RFM analysis, K-means, SOM, THSOM and association rules. In the practical part I describe appliacation of selected models to data from Acquire Valued Shoppers Challenge saved in Vertica database. I demonstrate usage of the models and their benefits. I apply cluster analysis on results of individual models to find clusters of similar customers and I also focus on the description of identified clusters.