Doctoral Thesis Proposal Defences in 2013

Ing. Gulshan Dovudov

Title: Tools and Resources for Computational Processing of Tajik Language
Supervisor: doc. PhDr. Karel Pala, CSc., FI MU
Opponents: doc. RNDr. Aleš Horák, Ph.D., FI MU
Prof. Zafar Djuraevich Usmanov, Institut matematiky AN RT, Tádžikistán
Date of the defence: 24th May 2013
Field: Computer Science and Technology - in English

Summary of the thesis:

The Tajik language is a variant of the Persian language spoken mainly in Tajikistan, where it plays a role of the national and official language for more than 7 million people. Since the Tajik internet society and consequently the potential market is rather small and Tajikistan itself is ranked among developing countries, available tools and resources for computational processing of Tajik as well as publications in the field are rather scarce. The aim of the thesis is thus to create some tools and resources which are still missing and which will be necessary for further more advanced processing of Tajik language. This work has two main goals: to develop fast Tajik morphological analyzer with a high coverage of real texts and to build a morphologically tagged corpus of Tajik which will be large enough for many empirical strategies for learning about language including a possibility of computation of basic word sketches for Tajik words, i.\,e. a corpus with at least 100 million words.
The thesis was defended.

Mgr. Andrej Gardoň

Title: Improving QA Systems with Temporal Reasoning and Multilingual Input
Supervisor: doc. RNDr. Aleš Horák, Ph.D., FI MU
Opponents: doc. RNDr. Marie Duží, CSc., FEI VŠB TU v Ostravě
Ing. Julius Štuller, CSc., Ústav informatiky AV ČR Praha
Date of the defence: 24th May 2013
Field: Computer Science and Technology

Summary of the thesis:

Question Answering (QA) systems are an emerging technology aiming at human-like interaction between a computer and a user. They process raw natural language (NL) sentences and provide a user with outputs having the form of natural language sentences.
This thesis proposal outlines a QA system (called Dolphin) based on Transparent Intensional Logic (TIL) that provides a computational model of natural language meaning. Dolphin enhances the QA field by the advanced temporal model that supports majority of the temporal phenomena including grammatical tenses, unspecified time intervals and relative time points. Dolphin as a system based on TIL, uses an inference module that keeps the system consistent and generates new knowledge from the stored one. Inference rules for the temporal reasoning are also provided. Dolphin is the multilingual system and for the purposes of the thesis it supports the Czech language and the controlled English language.
After the first introductory section, motivation to build Dolphin is presented in Chapter 2. In chapter 3 the overall architecture of QA systems is outlined.
This chapter also provides an introduction to the common components found in QA systems. Chapter 4 introduces well-known QA systems, and Chapter 5 reviews their abilities to process the temporal aspects. Chapter 6 resumes my work on the previous versions of the Dolphin QA system and it provides overview of my publications. The final chapter describes the aims of my thesis and a time schedule of the work.
The thesis was defended.

Mgr. Jan Rygl

Title: Determining Authorship of Anonymous Texts
Supervisor: doc. RNDr. Aleš Horák, Ph.D., FI MU
Opponents: Prof. dr. Walter Daelemans, University of Antwerp
prof. Ing. Karel Ježek, CSc., ZČU v Plzni
Mgr. et Mgr. Tatiana Tkačuková, Ph.D., PdF MU
Date of the defence: 24th May 2013
Field: Computer Science and Technology

Summary of the thesis:

In the authorship recognition task, texts are analysed to determine who is their real author. Disputed authorships of historical documents and literary works have been explored by many research studies since the 18th century and an authorship attribution of anonymous documents has played an important role in many criminal investigations. Over the past few years, authorship recognition techniques have been developed significantly, taking advantage of research advances in natural language processing and machine learning (ML).
The work is aimed at improving the ML approach and developing new authors' characteristics usable as attributes for ML:
1. A two-layer ML technique is proposed. Heuristic algorithms for generating attribute values for ML are replaced by the first layer of ML.
2. We replace similarities between an author and a document (according to the authors' characteristics) with the author's positions in rankings (generated from these similarities).
3. 3. A syntactic authors' characteristic is implemented and evaluated. The method is based on the syntactic analyser SET.
We not only focus on the authorship recognition task but also the process of collecting documents from the Internet. Intelligent methods of web monitoring and crawling are explored and utilized to build a corpus of Czech documents.
We are interested mainly in processing Czech but most of the proposed techniques are relevant for other languages as well.
The thesis was defended.

Mgr. Filip Mravec

Title: Numerical Processing of Spectrometric Data
Supervisor: prof. Ing. Václav Přenosil, CSc., FI MU
Opponents: RNDr. Zdeněk Kopecký, Dr., VF, a.s. Černá Hora
prof. Mgr. Tomáš Tyc, Ph.D., PřF MU
Date of the defence: 22nd May 2013
Field: Informatics

Summary of the thesis:

Neutron spectrometry is a scientific field that deals with monitoring of neutron radiation. This type of radiation is emitted during various nuclear reactions. Analysis of this radiation is important in many areas, especially in nuclear power engineering. One of the useful physical quantities is so called neutron spectral flux density, also known as neutron spectrum. As digital technologies become faster and easily available it is possible to effectively process data obtained from a detector in neutron-gamma mixed field. Many steps are required to process data from the output of a detector to the final calculation of neutron spectra.
In my thesis I want to focus primarily on three interconnected areas related to the processing of spectrometric data. Generally it is a development of suitable methods and algorithms for the calculation of neutron spectra. First topic is the calculation of so called response functions using Monte Carlo approach. Response function formally characterizes a detector and is different for various kinds of detectors. It is one of the necessary inputs for the spectrum calculation. Second part is a development of a support for energetic overlap of spectra. Various detectors are sensitive to different particle energies. It is beneficial to offer smooth overlap of spectra calculated in parallel in order to cover wider interval of energies. The last part is modification of current tools and development of missing tools in order to enable so called online monitoring of spectra, i.e. the data from the detector output are immediately used for the spectrum calculation without any intermediate offline steps. This would allow us to determine some radiation properties on the spot of the measurement.
The thesis was defended.

Mgr. Michal Růžička

Title: Maths Information Retrieval for Digital Libraries
Supervisor: doc. RNDr. Petr Sojka, Ph.D., FI MU
Opponents: doc. RNDr. Vlastislav Dohnal, Ph.D., FI MU
Prof. Michael Kohlhase, Computer Science - Jacobs University
Date of the defence: 22nd May 2013
Field: Computer Science and Technology

Summary of the thesis:

The usefulness of digital mathematics libraries depends on their ability to provide the user with all its content that they could possibly be interested in. However, sophisticated technologies and techniques developed for ‘textual’ digital libraries are not fully suitable for the specific needs of mathematical contents in digital mathematics libraries. Important missing feature is a robust implementation of mathematically aware full text search.
I aim to utilize existence of Math Indexer and Searcher (MIaS) system and underlying data set to develop, implement and evaluate techniques for the improvement of a mathematically aware search system for digital mathematics libraries.

The first steps were taken towards the improvement of MathML similarity search by the canonicalization of Presentation MathML: the next will be taken towards Content MathML normalization. Special attention will be paid to opportunities to exploit the context during the search in order to improve relevance of results and give feedback to users suggesting possible modifications to their query.
Further improvement can be achieved through the involvement of computer algebra systems in maths equality search. Optical character recognition processing of old publications led us to the idea of experimenting with searches for similar mathematical formulae based on the similarity of images of the printed formulae.
Evaluation of the implementations will be done on significantly large collection of real data. The results will be documented in the text of my thesis and published in peer reviewed international forums.
The thesis was defended.

Mgr. Daniel Tovarňák

Title: Towards Distributed Event-driven Monitoring Architecture
Supervisor: doc. RNDr. Tomáš Pitner, Ph.D., FI MU
Opponents: RNDr. Aleš Křenek, Ph.D., ÚVT MU
Prof. Dr. Dana Petcu, West University of Timisoara
Date of the defence: 22nd May 2013
Field: Computer Science and Technology

Summary of the thesis:

When monitoring distributed infrastructure, monitoring data related to the particular resource (e.g. Grid job, database, and the infrastructure itself) are typically produced by multiple distributed producers spread across many individual computing nodes. In order to determine the state and behavior of a resource all the relevant data must be collected, processed and evaluated without overloading the computing resources and flooding the network. The volume, velocity, and variability of monitoring data produced by modern cloud datacenters multiply and there is need for new approaches. Our goal is to propose a detailed design of real-time distributed monitoring architecture utilizing novel approaches and algorithms that will enable multiple simultaneous consumers to collect, process, and analyze monitoring data related to the behavior and state of many distributed entities.
The thesis was defended.

Mgr. Jakub Valčík

Title: Human Movement Models for Similarity Search
Supervisor: prof. Ing. Pavel Zezula, CSc., FI MU
Opponents: Mgr. Jakub Lokoč, Ph.D., MFF UK v Praze
Ing. Mgr. Zdeněk Říha, Ph.D., FI MU
Date of the defence: 22nd May 2013
Field: Informatics

Summary of the thesis:

Nowadays, public places are under surveillance by large amounts of video cameras and it is impossible to manually control each captured video for special interest activities. This dissertation proposal aims at utilizing such videos to automatically identify persons and recognize human activities from their movement. The human movement is represented as a sequence of poses obtained from captured video frames. Individual poses are used to extract characteristic features, such as rotations of joints angle, distances between couple of joints, etc. The extracted characteristic features along with a proper distance function form a human movement similarity model. The main objective is to design suitable models for classifying movement types and identifying people according to their gait. To achieve this objective we give an overview of state of the art approaches of activity and gait recognition with emphasis on existing similarity models. The gait recognition methods are categorized and exemplar characteristic techniques are provided. Each method describes feature extraction techniques, the way of their comparison and results of experimental evaluation. Our contribution to this topic, which has been already published, is described in the rest of this work.
The thesis was defended.

Mgr. Petr Bauch

Title: Control Explicit – Data Symbolic Model Checking
Supervisor: doc. RNDr. Jiří Barnat, Ph.D., FI MU
Opponents: doc. RNDr. Jan Strejček, Ph.D., FI MU
prof. Ing. Tomáš Vojnar, Ph.D., FIT VUT v Brně
Date of the defence: 21st May 2013
Field: Computer Science and Technology

Summary of the thesis:

Neither symbolic execution nor model checking in their present states are capable of verifying real-world parallel programs against temporal specifications. The combination of complete Peano arithmetic on integer variables and complex execution that must account for possible interleaving of parallel threads forms an insurmountable obstacle to state-of-the-art verification procedures. A considerable body of past research extended the verification community towards partial results (abstracting some aspects of the program, falsifying instead of verifying, etc.), yet complete verification of real-code was rarely considered.
The proposed PhD thesis aims at investigating the potential for achieving complete verification of a combination of symbolic and explicit approaches to model checking. Data-flow nondeterminism will be represented and modified symbolically and control-flow nondeterminism explicitly. The explicit state model checker DiVinE will be extended to support state space consisting of multi-states, instead of individual states. These multi-states comprise a set of different evaluation of input variables. The representation of these multi-states will be handled by an SMT solver, which will also decide the satisfiability of arithmetic and Boolean operations performed on the multi-states. The theory of bit-vectors appears to be of sufficient expressive power to support complete and precise verification, while still remaining efficiently decidable. Thus, the proposed investigation will concentrate on the limitations of the cooperation between DiVinE and the theory of bit-vectors.
The thesis was defended.

Mgr. Filip Jurnečka

Title: Key management schemes in wireless sensor network simulations
Supervisor: prof. RNDr. Václav Matyáš, M.Sc., Ph.D., FI MU
Opponents: Prof. Javier Lopez, University of Malaga
Prof. Rei Safavi-Naini, University of Calgary
Date of the defence: 21st May 2013
Field: Computer Science and Technology

Summary of the thesis:

In this thesis proposal we examine the field of key management schemes for wireless sensor networks. We investigate a large number of schemes, their classifications and evaluation possibilities. We identify several possibilities for improvement in both classification and evaluation of these schemes. Afterwards, we present our results in this area. Last but not least, we outline our future work proposal focusing mainly on enhancing current methodology for key management scheme evaluation. In addition, we will improve an already selected simulation tool for wireless sensor network system design and evaluation by adding the key management functionality together with a set of selected schemes.
The thesis was defended.

Mgr. Martin Stehlík

Title: Intrusion detection and optimization in wireless sensor networks
Supervisor: prof. RNDr. Václav Matyáš, M.Sc., Ph.D., FI MU
Opponents: doc. RNDr. Eva Hladká, Ph.D., FI MU
prof. Ing. Lukáš Sekanina, Ph.D., FIT VUT v Brně
Date of the defence: 21st May 2013
Field: Informatics

Summary of the thesis:

Wireless sensor networks (WSNs) are a type of ad hoc wireless networks with several specifics. A sensor node that is a building block of the WSNs is a low-cost device highly restricted in its resources regarding to the size of memory, the microcontroller performance and last but not least the energy supply limited by the battery capacity. The sensor nodes after the deployment are usually not managed by a human user; they can be controlled only through a base station (BS) and their life ends at the moment of the total battery depletion. Hence, the WSN has to be deployed with respect to these restrictions. The concept of the WSNs opens an area for novel attacks. We believe that at least some of them can be detected by an intrusion detection system (IDS) running on the sensor nodes in a distributive and collaborative manner. Since the IDS necessary brings an overhead to the computation and wireless communication of the sensor nodes, it should be deployed with respect to the restricted WSN. Our goal is to implement an IDS optimization framework based on proposal and work of Andriy Stetsko and Vashek Matyas. In this work we introduce the WSNs, compare them with “ordinary” wireless ad hoc networks and present the importance of the IDS and optimization of the detection techniques. We survey the attacks existing in the field of the WSNs, classify the IDSs, present available simulators usable for optimization of the WSNs and finally focus on optimization issues in WSNs with an emphasis on evolutionary algorithms. Finally, we present the planned work towards working IDS optimization framework incorporating suitable optimization algorithms that will optimize various detection techniques working in different environments.
The thesis was defended.

Mgr. Mária Svoreňová

Title: Formal Methods in Optimal Discrete Path Planning with Uncertainty
Supervisor: prof. RNDr. Ivana Černá, CSc., FI MU
Opponents: RNDr. Nikola Beneš, Ph.D., FI MU
Assist. Prof. Sertac Karaman, Massachusetts Institute of Technology
Date of the defence: 21st May 2013
Field: Informatics

Summary of the thesis:

Formal methods such as model checking and games on graphs provide a powerful mathematical tool in design and control of computer systems. In the last few years, they have been successfully employed in the area of robotics, for example to solve discrete path planning problems. In such a problem, the aim is to synthesize an optimal path in a discrete model of the robotic system subject to a given specification. Due to the use of sensors, the discrete model often involves some kind of uncertainty such as finite-horizon observation or partial observation. The focus of the PhD thesis is on the use of games and automata-based model checking techniques in discrete path planning problems for systems with uncertainty. The main goal is to develop new, optimal algorithms to solve such problems. For problems that are undecidable, such as LTL control for POMDPs, we aim to design efficient heuristics for well-defined subclasses of models or specifications. Prototype implementations and experimental evaluation will be used to demonstrate usability of the developed frameworks.
The thesis was defended.

Mgr. Sven Dražan

Title: Robustness of Stochastic Biochemical Systems
Supervisor: prof. RNDr. Luboš Brim, CSc., FI MU
Opponents: Prof. Paolo Ballarini, Ecole Centrale Paris, Laboratoire MAS, France
prof. Ing. Lukáš Sekanina, Ph.D., FIT VUT v Brně
Date of the defence: 9th January 2013
Field: Informatics

Summary of the thesis:

The aim of Systems biology is to study and understand the emergent behavior of complex biological systems composed of interacting entities such as molecules or cells by modeling, simulation, analysis and experimental evaluation. Evolution has rendered many functions of biological systems robust against changes of environment, therefore the understanding of how robustness emerges is essential for true understanding of a system’s function. Robustness according to Kitano is the cumulative measure of a system’s function being preserved across a set of external and internal parameter perturbations. Systems composed of interacting molecules such as signaling networks inside cells display stochastic behavior and can by thus conveniently modeled by continuous time Markov chains - CTMCs. The properties of a CTMC may be expressed in temporal logics such as Continuous Stochastic Logic (CSL) and analyzed using model checking, which can compute the probability of the property being satisfied in the CTMC model of the stochastic system.
The proposed aims of the thesis are: to define robustness of properties of CTMCs expressed in CSL over sets of parameter perturbations, to develop efficient algorithmic procedures for the evaluation of robustness, to prove the correctness of the algorithms with respect to the definitions, to implement the methods as a prototype tool and analyze the complexity on case studies. Due to the expected computational demands, the effective parallelization and/or distribution of the algorithms is also an aim.
The thesis was defended.

Mgr. Tomáš Majtner

Title: Analyzing Fluorescence Microscopic Images Through Texture Descriptors
Supervisor: prof. RNDr. Michal Kozubek, Ph.D., FI MU
Opponents: Prof. Cham Athwal, Birmingham City University, UK
Ing. Tomáš Suk, CSc., ÚTIA AV ČR
Date of the defence: 9th January 2013
Field: Computer Science and Technology

Summary of the thesis:

Digital images play an important role nowadays, with a big progress in last few decades. They have the ability to capture immediate moment from the different image domains of the real world. In this thesis proposal, we will concentrate our effort on one particular domain called fluorescence microscopy. We would like to recognize and classify images from this area. This can be done by using so-called image descriptors. The reason why we use image descriptors, when it is an easy task for biologist to recognize between different types of cells, is based on the fact that this task becomes harder when the number of images grows. A properly trained classifier with fast calculation of descriptors can be much faster than a skilled expert. Also the average occurrence of the mistakes is approximately the same in large collections. Another advantage of image descriptors is their ability to work with images of any dimensionality. For biologists, it can be hard to properly imagine and examine images of dimensionality higher than two. In this work, we introduce the current state of the art in the field with focus on global image descriptors, whose main characteristic is using of the whole image to extract the feature vector. HEp-2 cell classifier based on these descriptors is introduced as one of our results. In addition, we study the possibility of describing not only fluorescence images in 2D, but also in higher dimensions. Therefore, we present an extension of commonly used image descriptors called Tamura features for 3D images. Our aim for further research is to study selected image descriptors and suggest their improvements with focus on image classification.
The thesis was defended.

Mgr. Roman Stoklasa

Title: Semantic Segmentation Based On Similarity
Supervisor: prof. RNDr. Michal Kozubek, Ph.D., FI MU
Opponents: Prof. Cham Athwal, Birmingham City University, UK
Ing. Tomáš Suk, CSc., ÚTIA AV ČR
Date of the defence: 9th January 2013
Field: Informatics

Summary of the thesis:

In recent years, there is a rapid growth of multimedia and image data. Because of this, a problem of effective usage of such type of data has emerged. Problem of image recognition and understanding is one of the biggest unsolved problems in image processing field. Unfortunately, computers are not able to interpret image data, so many different approaches were developed to obtain at least some ability of image understanding. In our work we address the problem of semantic segmentation. Semantic segmentation is a process of division image into regions, where each region is assigned its semantics. This problem combines two different subproblems: image segmentation and object recognition. Object recognition can be done using image classification. In this work we summarize different approaches of segmentation algorithms, popular visual image descriptors and commonly used classification methods. We propose a combination of automatic image segmentation approach and image classification using kNN classifier for the purpose of object recognition in the image. We also show our two main results: (i) HEp-2 cell classifier and (ii) road detection algorithm, which combines image segmentation and image classification together. Our goals for further research are to develop common principles of semantic segmentation algorithm, which could be used in 3 different image domains: 1) in road detection algorithm, 2) for biomedical images and 3) for real-world images.
The thesis was defended.

Mgr. Tomáš Golembiovský

Title: Modelling of Thin Structures by Shell Elements
Supervisor: prof. RNDr. Luděk Matyska, CSc., FI MU
Opponents: RNDr. Igor Peterlík, Ph.D., IHU Strasbourg
doc. Ing. Jiří Sochor, CSc., FI MU
Date of the defence: 8th January 2013
Field: Computer Science and Technology

Summary of the thesis:

Objects with thin structures are present in many forms in our every-day life. Not only, for example, in the form of textiles, paper, wooden boards, leaves, tubu-lar structures but also in many anatomical or pathological structures (blood ves-sels, aneurysm, skin, …). Deformation modelling of thin-walled objects forms a specific area of research because the thinness leads to stability issues if methods for volumetric deformation modelling are applied.
The popular models (mass-springs, bending models) for modelling of such structures in real-time simulations give adequate approximation of the behaviour for applications in computer animation or cloth modelling. However, since the parameters of the models often do not have a clear meaning and the models de-pend strongly on the discretization they are not applicable to the areas where more physically-based approach is necessary (e.g. the surgical simulators).
Finite element methods have become popular in deformation of solids but have not yet seen much attention in the are of modelling of thin structures. This may be attributed to increased computational complexity compared to simpler models (mass-springs, bending models). This is indeed true, if high number of elements is used in discretization. We may, however, observer that complex dis-cretization is not always necessary for good approximation of the simulated be-haviour. On the other hand, high number of elements describing properly the geometry of the surface is required for visualisation, contact modelling or inter-actions. We propose the use of FEM model using shell elements. Exploiting the bending property of the shell elements we employ two meshes for the simula-tion: low resolution mesh for mechanical model and high polygonal mesh that is mapped onto the geometry of the shell elements for other tasks.
The main objectives of the proposed thesis can be divided into the following three topic: First, formulation of shell element based on surface of higher order is necessary for efficient simulation of curved geometry of thin structures. Second, techniques for mesh construction and mesh adaptivity that can take advantage of the underlaying high-order surface need to be defined. Proper exploitation of the bending property allows us to use much lower number (in order of magnitude) elements during the simulation. And finally efficient implementation of the de-signed algorithms for modern GPGPUS can provide significant speedup that is so much needed for real-time simulations.
The thesis was defended.

Mgr. Martin Komenda

Title: Advanced system for optimization of medical curriculum
Supervisor: prof. RNDr. Jiří Hřebíček, CSc., PřF MU
Opponents: doc. Ing. Josef Hanuš, CSc., UK v Praze, LF v Hradci Králové
doc. Mgr. Hana Rudová, Ph.D., FI MU
Date of the defence: 8th January 2013
Field: Computer Science and Technology

Summary of the thesis:

In the Thesis a brand new and original medical curriculum optimization methodology within tertiary education will be described by adopting an outcome-based approach and applying modern IT and communication technologies. Existing solutions that have been published are focused on the curriculum only from a certain perspective, offering the agenda together with selected functionalities and making an effort to provide them to students and teachers of the respective institution in a transparent format. However, there still does not exist a complex instrument that would cover all elements connected with global curriculum optimization, including a detailed parametric description down to the level of learning units, and that would be linked to learning objects. The aim of the Thesis is to create a new web-oriented tool included advanced methodology, which provides optimization of medical curriculum using learning outcome approach.
The goal of the research is focused on three topics:
• To propose an extended methodology which works on existing and published results, and which will help optimize curriculum structure in medical fields of study.
• To develop a web-oriented platform based on described methodology for complex curriculum content management and visualization.
• To analyze data which will be collected during pilot optimization process of medical curriculum.
The thesis was defended.

Mgr. Marek Vinkler

Title: Acceleration Data Structure Construction for Ray tracing
Supervisor: doc. Ing. Jiří Sochor, CSc., FI MU
Opponents: Ing. Jiří Bittner, Ph.D., KPGI FEL ČVUT
RNDr. Pavel Matula, Ph.D., FI MU
Date of the defence: 8th January 2013
Field: Informatics

Summary of the thesis:

The basic task of computer graphics is to synthetize a two-dimensional image from the input scene description which is mostly three-dimensional. A special case of this task is photorealistic rendering which computes the resulting image by physically-based methods. Photorealistic visualization is used in a variety of industries ranging from scientific visualization, computer-aided design in car industry or architectural visualizations to the creation of computer generated films. Unluckily, rendering of photorealistic images is a time consuming task and is thus realized on powerful rendering farms. Recently acceleration of this class of algorithms to a level enabling it to run on a commodity desktop computer has been a hot research topic. Such acceleration would not only reduce the visualization expenses in areas such as engineering or film industries but it would also enable photorealistic rendering to be used in new areas such as the video game industry. Most methods capable of photorealistic rendering use the ray tracing algorithm to compute the visibility inside a virtual scene, and this algorithm is the most time consuming part of the entire image computation.
The contribution of this work is the design, implementation and evaluation of methods for ray tracing acceleration. In this quite broad area this work focuses on the construction of data structures for fast spatial searching in the rendered virtual scenes. Construction of these structures can be optimized either for the build time or the quality of the spatial sorting, both of these parameters have been researched in this work and novel solutions are proposed for them. Given the highly parallel nature of the ray tracing this work is focused on parallel algorithms running on modern graphics chips.
The thesis was defended.

Mgr. Hana Bydžovská

Title: Course Enrolment Recommender System
Supervisor: doc. Ing. Michal Brandejs, CSc., FI MU
Opponents: doc. RNDr. Vlastislav Dohnal, Ph.D., FI MU
doc. Ing. Kristína Machová, Ph.D., Technická univerzita v Košiciach
Date of the defence: 7th January 2013
Field: Computer Science and Technology

Summary of the thesis:

The proposed thesis will design and evaluate appropriate recommendation techniques to select courses that are beneficial or interesting for students. These techniques will be implemented to the data mining tool Excalibur. This tool is suitable for analytical data processing. Excalibur contains a data warehouse that stores study-related data, including its history. The aim of the thesis is also to integrate the recommender system into Excalibur. This module will be used to recommend courses to students with respect to the study success. The enhanced version of Excalibur will be a part of the Information System of Masaryk University.
The thesis was defended.

Mgr. Daniel Jakubík

Title: Exploratory Search in Digital Libraries
Supervisor: doc. Ing. Michal Brandejs, CSc., FI MU
Opponents: RNDr. Miroslav Bartošek, CSc., ÚVT MU
doc. RNDr. Vlastislav Dohnal, Ph.D., FI MU
Ing. Martin Svoboda, Národní technická knihovna, Praha
Date of the defence: 7th January 2013
Field: Computer Science and Technology

Summary of the thesis:

Digital libraries have proven to be an important part of the scientific communication model. This has been achieved due to their ability of long term preservation and presentation of papers and other scientific works. Unfortunately, with the increasing amount of data it becomes more and more challenging to utilise their epistemic potential.
Within the Masaryk University Information System, we have recently run the institutional repository. On the same platform we have build new system Repozitar.cz in cooperation with 15 other universities. The aim of the system is an inter-university exchange of scientific outputs.
The aim of the PhD thesis is to suggest and evaluate suitable techniques of exploratory search which can be applied to rich content of digital libraries. The special emphasis will be placed on searching that will respect the level of current user knowledge. Instead of traditional searching using keywords, the query will consist of the set of papers and their relations. The proposal of an interactive visual interface will be an integral part of the thesis. It will also focus on aspects of practical implementation and usage in both systems mentioned above.
The thesis was defended.

Mgr. Bc. Petr Novotný

Title: Modeling and Verification of Infinite-State Reactive Systems
Supervisor: prof. RNDr. Antonín Kučera, Ph.D., FI MU
Opponents: Asst. Prof. Krishnendu Chatterjee, Ph.D., IST Austria
prof. RNDr. Mojmír Křetínský, CSc., FI MU
Date of the defence: 7th January 2013
Field: Informatics

Summary of the thesis:

The thesis proposal contains new results from the area of an algorithmic analysis of infinite-state stochastic graph games. Such games can be used to model various discrete reactive systems, i.e., systems that in some way react to the inputs from the user or from the environment. The thesis proposal covers two main topics: the determinacy of general infinite-state graph games with unbounded payoff functions; and an efficient algorithmic analysis of games played over transition graphs of finite-state machines equipped with one or more integer-valued counters. Apart from other results, we introduce a new special class of games over these counter machines, so called consumption games, that can be used to model various reactive resource-dependent systems. The aim of the Ph.D. thesis, that is presented in this thesis proposal, is to extend these results even more and present all the relevant results in a unified manner. The design of efficient algorithms for the analysis of games over counter machines is our primary task, with a special attention being given to the formalism of consumption games.
The thesis was defended.

Mgr. Vít Rusňák

Title: Interaction Methods for Large High-Resolution Screens
Supervisor: prof. RNDr. Luděk Matyska, CSc., FI MU
Opponents: prof. Jason Leigh, University of Illinois, Chicago
prof. Ing. Miroslav Švéda, CSc., FIT VUT v Brně
Date of the defence: 7th January 2013
Field: Computer Science and Technology

Summary of the thesis:

A dramatic onset of multi-touch interfaces has been observed in recent years. Besides consumer electronics, such as smartphones, tablets or personal computers, there is a category of horizontal (tabletops) and vertical (display walls) interactive systems. Although multi-touch sensors generally provide multi-user capability, they are not able to associate input events with individual user. Providing methods for assembling various kinds of sensors (e.g., multi-touch panel and web camera) we can develop complex user-aware interactive environments for the next generation of group collaborative systems.
In this work, we present various interaction techniques used in tabletops and interactive display walls with focus on user distinguishing and their association with input events. The aim of the thesis is a research and development of techniques which provide context-rich personalised description of input actions. The techniques will be generally suitable for systems based on commodity hardware which integrate multiple form-factor sensor devices. These can be arbitrarily connected to multiple computers that communicate over a local network and processing of input data can be spread across such distributed environment.
The thesis was defended.