List of Theses Defences in the Year 2008

Mgr. Marek Grác

Summary of the thesis:

Automatic translation of natural languages, often also called machine translation is among the most challenging problems in IT. Universal machine translator would have huge social, political and economical impact. In my work, I am focussing on methods suitable for translation of very close languages, eg. Czech and Slovak. Translation among those languages, although is a lot simpler than for more distant languages, but the expectations of educated users are significantly higher. The goal of my work is to develop fully- and semi-automatic methods for localisation of language resources, which are currently not available for the other language. The data obtained this way will be used for improving quality of machine translation system based on surface syntax analysis.
The thesis was defended.

Mgr. Karel Kisza

Summary of the thesis:

The thesis was defended.

Mgr. Matěj Štefaník

The thesis was not defended.

Mgr. Petr Beneš

Summary of the thesis:

Computation of tunnels in protein molecules is an important area in protein analysis. It is crucial to perform complex analysis of tunnels connecting the specified cavity inside the protein molecule with the surface of the molecule. Various techniques, mainly based on the Voronoi diagram and Delaunay triangulation, have been proposed to analyze tunnels in static molecules. However, protein molecules are dynamic structures and atoms change their positions in time and with this movement the tunnels also change -- widening and narrowing of tunnels can be observed. The analysis has to cover the behaviour and stability of certain tunnels. The aim of the thesis is to present new techniques for tunnel computation and tunnel analysis in order for chemists to be able to judge on tunnel relevance. The methods will be mainly targeted on large dynamic sequences of protein molecules and will be based on computational geometry principles and tests on real biochemical data will be included.
The thesis was defended.

Mgr. Ivan Fialík

Summary of the thesis:

Quantum information processing is a very interesting and important field which overlaps with physics, mathematics and informatics. It deals with harnessing laws and phenomena of the quantum world for computation. It especially studies what they enable us to do which goes beyond the abilities of classical information processing. Communication complexity of some distributed computational problems can be reduced with the help of quantum information processing. We speak of pseudo-telepathy when it is able to completely eliminate the need for communication. Such problems are often described using a terminology of the game theory and they are usually called pseudo-telepathy games. Apart from the fact that they can be seen as distributed problems which can be solved using quantum entanglement without any form of direct communication between the parties, there is one more reason to be interested in pseudo-telepathy games. They offer an alternative way to show that the physical world is not local realistic. The thesis will be focused on the problem of finding the smallest size of inputs for which there is no classical winning strategy for a given pseudo-telepathy game and on the problem of finding an upper bound for the success of the best possible classical strategy. Next goal is to examine the impact of imperfections occurring during an experimental implementation of a quantum winning strategy on the success of quantum players.
The thesis was defended.

Mgr. Jan Sedmidubský

Summary of the thesis:

Due to the exponential growth of digital data and its complexity (like multimedia, text documents or scientific data), we need a technique which allows us to search such collections efficiently. A suitable solution is based on distributed systems and the metric-space model of similarity. However, the biggest challenge tends to shift towards self-organized systems. Unlike distributed access structures, self-organized networks are capable of adapting to environment's changes and allow a network to correct any damage with minimal human interaction. In general, such types of systems are able to control and evolve autonomously. Many researchers have currently been attracted by advantages of social networks and have proposed many systems combining the peer-to-peer communication paradigm and social-network principles. The survey part of this proposal describes such types of systems and divides them into four different areas according to their application domain. Because such approaches exploit diverse techniques of searching and maintaining own structure, we compare and group them by their common properties into several categories. Based on the study of the current state, we have recognized the need for a new self-organized structure for similarity searching exploiting social-network paradigm. We propose a self-organized network, called Metric Social Network (mSN), where peers are linked directly if they are semantically similar with respect to a certain query. Ties among peers are constructed from the answers returned to queries. In particular, each peer maintains a list of queries it asked or answered, called the query history. The query history represents the peer's knowledge about the network. It is exploited by a navigation algorithm in order to route a query efficiently. Preliminary experimental results indicate a significant boost in searching performance.
The thesis was defended.

Mgr. Dalibor Klusáček

Summary of the thesis:

This thesis proposal is focusing on the are of Grid scheduling. Grid is distributed and decentralized computer environment composed of large number of heterogeneous resources interconnected by high speed communication network and managed by different owners and providing various services for its users. Grid environment is changing dynamicaly in time. Job scheduling in such environment is not an easy task since it has to react properly to various dynamic events such as job arrivals, job completions or resource failures and startups. Moreover it is very important to guarantee nontrivial Quality of Service (QoS) for the users together with good resource utilisation so that also resource providers are satisfied. Current scheduling techniques applied in the Grids are mostly based on the queueing systems. While single objectives can often be satisfied with some simple queue-based policies, complex requirements such as small response time, deadlines, resource utilisation, advanced reservations, etc., are hard to achieve by a queue-based solution, especially for common users. In our work we are focusing on the application of advanced scheduling methods based on schedule creation and optimisation with the goal to provide better performance than current state-of-art queue-based solutions. We use dispatching rules to create the initial schedule incrementally in time as new jobs arrive. Local search-based optimisation is then used to optimise or correct this initial schedule by performing local changes within existing schedule. Here the anytime approach is applied which allows us to quickly and properly react to changing situation. Proposed approach is driven by incoming events that represent the dynamic changes in the Grid environment. According to the type of the event either dispatching rule or local search or both are used to deal with the new situation. Nowadays, schedule-based approach is considered as slow solution especially for static cases and large problems. This work will study the suitability of schedule-based approach applied in the dynamic environment. Since the Grid dynamics introduces problems such as uncertainty, dispatching rules and local search will use event-based, incremental and anytime approach to keep the schedule valid through time using reasonable computation time. Then the performance and suitability of proposed schedule-based approach will be evaluated by various experiments in the simulated Grid environment.
The thesis was defended.

Mgr. Vojtěch Krmíček

Summary of the thesis:

Growing speeds of computer networks and rising number of Internet users increase requirements for effective network monitoring and surveillance and a lot of past methods are out of date. Widely used approach for high-speed monitoring is based on aggregation of IP flows. Such information provides no hint about the content of the transmitted data, but detailed information about network behavior and characteristics of network traffic. We can use with advantage specialized hardware based solutions developed for highspeed network monitoring. During Liberouter project at Masaryk University we have developed such hardware, namely hardware accelerated network probes for IP flow monitoring and for intrusion detection. The aim of my thesis is to study possibilities of monitoring and anomaly detection in high-speed network environment using hardware-accelerated devices. In theoretical part, I will study both existing and newly rising network threats, their characteristics and possibilities of detection with IP flow information. A practical part of my work will be focused on hardware accelerated high-speed network monitoring and on designing of new anomaly detection and prevention methods.
The thesis was defended.

Ing. Oladimeji Oniyide

The thesis was not defended.

Mgr. Šimon Řeřucha

Summary of the thesis:

Despite the swift development in the field HMI, the machine control remains solely a human task. The most common cause of faults in the operation of complex man-machine systems designed for high reliability is the failure of human factor, typically caused by fatigue and loss of attention. It is assumed that the best guidance for detection of such situation is brain activity, since the brain is responsible for the psycho-physiological state of a person. The seeking for the relation between the state of the operator and their level of vigilance is an aim of the field of Neuroiformatics. The typical representative of an operator is a vehicle driver. The common result of the driver failure is a traffic accident and the subseqent physical damage and casualties. We focus on the detection and prevention of the situations when the driver's vigilance drops below certain level. The principal idea is to monitor the psycho-physiological state of the driver and assess whether the driver is still competent enough to continue driving. We propose a model of a system that will make the assessment on the basis EEG signal processing, automated visual recognition and analysis of driving behaviour. The relation between the psycho-physiological state of human operators and their level of vigilance has been of interest to neurologists for several decades. There is a considerable amount of neurological data measured within a number of experiments within various research projects in this field. Since data acquisition is a complex and expensive process, there is an effort to exploit the existing data more effectively. We present our approach that is expected to result in a more effective utilization of such data. The principal idea is to propose a generic structure of such data and to develop tools that will allow the researchers to gather the data and build compact datasets with well defined criteria for acquisition and structure. This dataset is to be called the „Neuroinformatic Database“. The second step is to design an environment, that would allow the neuroinformatic community to share and interchange the content of their own neuroinformatic databases.
The thesis was defended.

Mgr. Václav Brožek

Summary of the thesis:

The study of stochastic 1 1/2- and 2 1/2-player games is motivated by the need for models of systems involving uncertainty as well as nondeterminism. Systems with controller or systems interacting with an unpredictable environment are examples of such systems. Moreover whenever a kind of recursion appears in the system, an infinite-state model is needed. Stochastic games played over the transition graphs of pushdown automata (PDA) are designed to be the appropriate model for such systems. Their study was initiated by the research on a special case---probabilistic pushdown automata. Recent results target already 1 1/2 player games and the reachability objective. The aim of the thesis is to continue this research by completing the results partially achieved and to extend it. In particular, decidability of variants of reachability problems for PDA games and their subclasses is among the targets, as well as model-checking with Büchi objectives and possible also some fragments of PCTL. Eventually, the thesis will try to explore the situation in 2 1/2 PDA games.
The thesis was defended.

Mgr. Vojtěch Forejt

Summary of the thesis:

We consider finite stochastic turn-based games of two and half players. These games can be used to model systems in which the exact behaviour is unknown or uncertain and can be expressed only by introducing probabilities of transitions among states. We study a subclass of games in which a winning objective is specified by formula of a branching time logic. The problem is to decide if a player can play so that the formula is satisfied. We present some known results, i.e. that the problem is undecidable in general, but is decidable when some restrictions are put on the structure of formula or game. The aim of the thesis is to extend the known results and to study other types of games (e.g. the games of incomplete information).
The thesis was defended.

Mgr. Jan Holeček

Summary of the thesis:

The World Wide Web has been studied since its emergence in 1990s. Link analysis, among other methods, proved very useful in searching and page classification tasks. Most link analysis algorithms can be formulated in terms of random walks on the Web graph and hence randomized models of users. The models being used are very simple, though. Recent advances in the field of probabilistic pushdown automata (pPDA) suggest that pPDA can effectively be used as a model of a user clicking on the Web. The intended thesis will extend results on pPDA and use them to define and effectively express characteristics of pages, e.g. page ranking. It will also compare the new methods with existing ones.
The thesis was defended.

Mgr. Filip Andres

Summary of the thesis:

The thesis was defended.