Ing. Mgr. Ladislava Smítková Janků, Ph.D.

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

prof. RNDr. Tomáš Skopal, Ph.D.

Department of Software Engineering

This thesis contributes to automated interior design research by designing and implementing a new system. The design process is divided into the planning of the functional zones using simulated annealing and arranging the furniture with a genetic algorithm. The system can fulfill the user's requirements on the room's shape, functions, and used furniture. The experiments show the ability of the system to design an interior of rooms with varying shapes and selection of furniture.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

doc. Ing. Kamil Dedecius, Ph.D.

Department of Applied Mathematics

Stock markets are vulnerable to various factors, and therefore their prediction is very challenging. There are many approaches to forecasting stock prices, but none is universal. This thesis focuses on an approach called machine learning. CNN, LSTM, GRU models, their bidirectional variants and model combinations can be included in this approach. Furthermore, the individual models are used for one-day and multi-day forecasting. The thesis points out the unusability of the one-day prediction, which is repeatedly used to predict several days ahead. SXR8.DE data, which replicates the S&P 500 index, was selected for the forecast. The models are then evaluated on this data, and based on their accuracy, the model with the highest success rate is selected in order to maximize the profit from stock trading.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

doc. RNDr. Ing. Marcel Jiřina, Ph.D.

Department of Applied Mathematics

This Bachelor's Thesis deals with the detection of extremism in the online environment, speci-\linebreak fically the detection of Neo-Nazi and Salafi websites. The aim is to create a hybrid knowledge system that will classify the given websites using a Support Vector Machines method and the classification rules of the knowledge system. The solution to this problem is based on thorough research of existing solutions for detecting extremist texts, images and videos and then their analysis with the help of expert literature. In this work, a dataset of extremist and non-extremist websites was created using existing text documents and images. The SVM method was used for text classification and its output served as input to the knowledge system. This was created using the PyKe shell and the whole program was programmed in Python. The SVM models gave outstanding results with around 99 \% classification accuracy. The overall classification accuracy of the system was 80 \%.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

doc. Ing. Kamil Dedecius, Ph.D.

Department of Applied Mathematics

This bachelor thesis deals with the simulation of artificial life. The simulation model is based on the principle of multi-agent system, meaning that each agent (organism) in the simulation is simulated independently of the others, but the agents can still interact with each other. The evolution of the agents is guided by evolutionary algorithms, more precisely genetic algorithms, using the principle of artificial genome that determines the characteristics and behaviour of the agent. The environment in which agents (or populations of agents) live is interactable and time-varying, placing greater demands on their ability to adapt. The entire program can be easily modified with a wide range of parameters, extending the actions that agents are allowed to perform and adding or modifying the biomes that instantiate the environment. The purpose of this work is to create a program capable of such simulation and to create a platform for future development of this program, i.e. the program must be easily extensible and independent of implementation modules.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

doc. RNDr. Ing. Marcel Jiřina, Ph.D.

Department of Applied Mathematics

This work aims to create an intelligent adaptive tutor system that will practice a selected area of artificial intelligence - decision trees. The desktop application is based on an initial review of existing adaptive tutoring systems, technologies for building expert systems, and then leveraging one of the existing technologies. The technology used is CLIPS JNI - the Clips language and its integration with Java and Swing for GUI implementation. A knowledge base is created with the help of an expert (thesis supervisor), based on which rules for the operation of the expert system are created. The resulting system is a diagnostic system that generates a sequence of questions based on the user's answers, where at the end it evaluates the user's knowledge and displays where the user went wrong. The user can choose a level and a domain, with the level varying in dependence on the correct/incorrect questions answered. The work can be used as a basis for further exploration of the area, but can also be extended to teach other areas of AI or other subjects, thus helping to make teaching more effective.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

doc. RNDr. Ing. Marcel Jiřina, Ph.D.

Department of Applied Mathematics

The focus of this thesis is recognition of extremist texts using supervised machine learning models. The goal is to create a classification system capable of specific extremist text detection. Three different datasets were created in this thesis, with the extremist one being focused mainly on nazi and neo-nazi texts. Proposed classification system creates predictions based on weighted votes of partial classificators. These classificators were created using an implemented training program. Appropriate combinations of methods for their creation were chosen based on performed experiments. In these experiments classification success rate was estimated based on used text preprocessing method and feature extraction method. Two classification models were used for text classification, specifically SVM and Random forest. Accuracy of the final classification system is estimated to be 85 \%.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

Mgr. Eva Pernecká, Ph.D.

Department of Applied Mathematics

This bachelor thesis builds upon contemporary works that deal with creating hotel simulators using simulation-based optimization and Monte Carlo methods. The literature review summarizes the contemporary use of mathematical and artificial intelligence based method in solving revenue management problems. An analysis of strengths and weaknesses of existing hotel simulators is performed, based on which we present suggestions for improvement of the current solution. A novel approach to modelling customer demand is presented, which is based on the concept of each customer choosing a distinct set of 'price walls' (prices with 0 % and 100 % probability of acceptance), rather than having a fixed perception of price, as implemented in previous works. Additionally, we discuss the possibilities of using additional types of price multipliers, as well as their utilization in altering the shape of the demand function. Our experiments have shown that the model behaves realistically and is capable of improving through optimization, achieving an increase of 3.43, 8.04 and 20.29 % in total revenue of three simulated hotels with the capacity of 10, 25 and~75 rooms. This observation suggests the viability of the model as an alternative to the existing solution for further research.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

Ing. Jan Glaser

Department of Applied Mathematics

This thesis aims to investigate the application of neural networks in a chess engine. Neural networks are trained using supervised and reinforcement learning. The supervised learning part of the thesis involves automatic dataset generation from the Stockfish chess engine. This part also includes an analysis of the relationship between the MiniMax algorithm and the output of a neural network's single forward pass, as well as the connection between the depth of the generated dataset and the use of the appropriate type of ResNet architecture. Proposes metrics of board complexity and reveals limitations of convolutional neural networks to find and utilize information hidden in the game tree. In addition, the thesis compares multiple approaches to reinforcement learning of the chess engine, proposes novel method to reinforcement learning, which achieved better results than standard method in two practical experiments.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

Dr. techn. Ing. Jan Legerský

Department of Applied Mathematics

The Cutting Stock Problem is a common optimization challenge in manufacturing and logistics, where the goal is to efficiently arrange objects or shapes within a larger container, in order to minimize waste or maximize utilization. This thesis aims to investigate the current existing solutions to the Cutting Stock Problem, and provide a potential solution to Cutting Stock problem involving irregular shapes, using genetic algorithm. The findings of this research have the potential to contribute to a more efficient and sustainable manufacturing processes of variety of industries such as clothes manufacturing, or automobile manufacturing. In doing so, I aim to provide an accurate assessment of viability of genetic algorithms for solving such problems, for the company I am currently working for.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

Ing. Jiří Novák, Ph.D.

Department of Software Engineering

This thesis describes the process of developing an interactive tool for visualization of the process of solving selected optimization problems using genetic algorithms. The theoretical part provides an introduction to the evolutionary algorithms, more specifically then focuses on genetic algorithms. This part also lists typical optimization problems that are commonly solved using genetic algorithms. The practical part describes all the important steps in creating new functional software. This includes the comparison of existing solutions, analysis of possibly suitable technologies and libraries, and software documentation. The implementation of the application, using the Blazor WebAssembly framework, is also described in this part. The interactive tool provides the visualizations of three typical optimization problems and the problem of 2D bitmap image generation. At the end, the thesis lists the techniques used for testing and documentation of the application. The final result of this thesis is a functional interactive web application.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

Ing. Monika Borkovcová, Ph.D.

Department of Applied Mathematics

A major challenge for modern machine hearing systems is the problem of sound source separation. One of the approaches to solve it lies in the studies in computational auditory scene analysis (CASA), in which computational models are inspired by the mechanisms in the human auditory system and the widely-known "cocktail party" effect. This thesis is meant to be an introduction to the field of CASA. In the first part, it investigates the theory behind auditory modeling and sound processing by diving into the physics of sound, biology of the human ear, psychology of hearing and digital signal processing. In the second part, this knowledge is practically used to implement a simple CASA system for separating monophonic music from background noise. This system is then extensively described and experimentally evaluated on a set of piano recordings.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

Mgr. Vojtěch Rybář

Department of Applied Mathematics

The subject of this thesis is the application of machine learning for automatic real estate object rent price prediction, based on data from Czech real estate listing websites. A dataset containing technical parameters of real estate objects is compiled, visualized, and used to train several machine learning models, which are then compared. Next, the thesis focuses on the usage of image data to improve real estate price prediction. The possibility of descriptive feature extraction from image data is explored. Methodologies for manual feature extraction are introduced, and several features, such as the level of natural light in the object, are experimentally extracted from image data. A dataset containing descriptive image features is compiled, visualized, and used with machine learning models. The inclusion of image data features was measured to decrease the error rate of the price prediction of the used models by approximately 10 to 20%.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

Mgr. Dalibor Šidlo

Department of Applied Mathematics

This thesis delves into the challenges of event-based ticket sales data, with the goal of enhancing data collection and proposing models for dynamic pricing. A literature review examines various approaches, while data analysis identifies key insights and obstacles in the dataset provided by a commercial partner. A refined data collection strategy, incorporating diverse data sources, is proposed alongside a synthetic data generation model to test out the proposed methodology. Although significant improvements were not observed, valuable insights into the potential areas for further refinement were gained. Proposed dynamic pricing models demonstrated good performance during low-demand periods, indicating their potential use for discounts and optimizing ticket pricing strategies. This research contributes to the field of event-based ticket sales by investigating different approaches, identifying challenges, and laying the groundwork for future advancements in effective and efficient pricing strategies in the field.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

prof. RNDr. Pavel Surynek, Ph.D.

Department of Applied Mathematics

This thesis is focused on developing decision trees for stock price prediction of Bitcoin and gold that use fundamental and technical market analysis.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

Department of Applied Mathematics

This bachelor's thesis delves into the utilization of neural networks in designing specific components of home interiors. The objective of this research is to extract features including color, style, and material, and employ these features to construct a graph neural network that generates combinations of patterns and accessories. The methodology adopted in this work entails the utilization of VGG for feature extraction, YOLOv7 for object detection, and GCN for compatibility prediction.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

Mgr. Petr Šimánek

Department of Applied Mathematics

This thesis is focused on the evaluation of the optimization framework SEAGE in the context of the current research state in the hyper-heuristic field. This is done using our new metric, which evaluates algorithms objectively according to the quality of the optimization problem instances. The comparison is based on comparing already implemented heuristics in the SEAGE and hyper-heuristics created for the international competition CHeSC2011. At the beginning of this thesis, we describe our motivation, goals and also the current state of hyper-heuristic research thanks to which we found out about the HyFlex framework together with the hyper-heuristics of the CHeSC2011 participants. During the implementation we encountered a series of problems that discouraged us from using the already implemented evaluator and metric in the HyFlex framework. We describe these problems in detail and present a new metric and evaluator, which solves mentioned problems and also brings a certain form of improvement into the SEAGE framework. For the introduction of hyper-heuristic creation in the SEAGE framework we've created one that uses the newly introduced improvement. For the testing of the newly introduced metric we first recalculate the results of the international competition CHeSC2011 and examine the behavior of algorithms from both the frameworks. An interesting output has emerged from these observations the quality of the hyper-heuristic building blocks (metaheuristics) solutions in the SEAGE framework for the problem domain SAT is really good. Finally, we summarize all the gained observations and based on the results from experiments we evaluate the SEAGE framework in the terms of the current state of hyper-heuristic research.

Thesis on DSpace

Ing. Mgr. Ladislava Smítková Janků, Ph.D.

prof. RNDr. Pavel Surynek, Ph.D.

Department of Applied Mathematics

This thesis addresses the problem of agent cooperation in finding and collecting objects in an unknown environment on a grid graph. First, it describes the currently used methods for dealing with the exploration problem and agents' decision-making patterns. Afterwards, a series of typical cases and a model that allow their implementation in a simulated environment is proposed. This work presents experiments for combinations of the chosen behaviors and modules for the agent in the scenarios created. Based on the results, the effects of the selected agent types will be determined.

Thesis on DSpace