Programme of colloquia with abstracts for the Autumn 2003 semester

23 September 2003

Doc. Ivan Kopecek, CSc, FI MUNI Brno

Divide et impera – speech segmentation

30 September 2003

Doc. Karel Pala, CSc, Faculty of Informatics, Masaryk University, Brno

A Czech corpus containing errors (proposal for a punctuation corrector)

Abstract: In this lecture, we shall focus on the issue of a text corpus containing errors, which was created at the NLP Lab, and analyse the errors within it (i.e. punctuation, morphological, syntactic and stylistic errors). We shall also discuss the method of tagging errors and the software tool used for this purpose. Finally, we shall attempt to outline what a punctuation corrector for the Czech language might look like.

7 October 2003

Prof. Miloš Dokulil CSc, Faculty of Informatics, Masaryk University, Brno

TheComputerisation of the Mind

14 October 2003

Prof. Jiří Wiedermann, DrSC, UI CAV Prague

‘New’ artificial intelligence through the eyes of a theoretical computer scientist.

Abstract: Towards the end of the 1990s, a new paradigm known as ‘embodied cognition’ began to gain ground in artificial intelligence; this can be summarised by the phrase ‘intelligence needs a body’. This shift from viewing cognitive computations purely as a data-processing process to processes governing the acquisition of such data through perception and motor skills marked a revolution not only in the understanding of the principles of artificial (and ultimately natural) intelligence, but also in their applications, particularly in humanoid robotics. The role of imitative learning – which is impossible without a body and sensory organs – is becoming increasingly recognised. In this regard, a significant biological discovery from the end of the last century stands out – the discovery of so-called mirror neurons in the primate brain. These neurons represent the fundamental mechanism for the realisation of imitative learning. In this lecture, we will provide an overview of the trends outlined here from the perspective of theoretical computer science. In particular, we shall focus on the computational model of the embodied agent and, within this framework, on the role of mirror neurons in sensorimotor coordination, imitative learning and the development of thought.

21 October 2003

Dr Alica Kelemenová, CSc, UI, University of Opava

Eco-grammatical systems

Abstract: Research into cooperative systems has also found its reflection in contemporary theory of formal grammars and languages. Typical examples of such systems include colonies, sequential and parallel grammatical systems, and ecogrammatical systems. We shall present current research trends in this field using the model of eco-grammatical systems. This is a model with a natural biological motivation (an environment containing organisms), developed on the basis of previous studies in the field of artificial life. In the lecture, we will characterise the behaviour of eco-grammatical systems, the computational power of the model, the influence of environmental and population characteristics on the development of the eco-grammatical system, and the phenomenon of emergence.

4 November 2003

Prof. Philippe Schnoebelen, LSV, ENS de Cachan, France

Formal verification of lossy channel systems

Abstract: Many asynchronous communication protocols are designed to cope successfully with unreliable communication links (e.g. the Internet). A natural model for such protocols is the Lossy Channel Systems introduced by Finkel, Abdulla and Jonsson. Paradoxically, it turns out that algorithmic verification is easier for lossy systems than for systems with reliable communication links! In this talk, we survey the main results on the verification of lossy channel systems and discuss some of the open problems we are currently interested in.

11 November 2003

Prof. Jaroslav Pokorny CSc, Faculty of Mathematics and Physics, Charles University, Prague

XML Databases – Current Status, Problems and Prospects

Abstract: The current increase in the number of documents expressed in XML and certain types of applications are driving the need for XML databases. XML can thus be understood as a new data model, and an XML database as a new phenomenon introducing new data structures, query languages, indexing mechanisms, and so on. XML databases, and in particular the implementation of the XQuery language, represent a challenge that has been taken up by a significant part of the theoretical database community as well as by the leading database vendors. In this lecture, we will outline the types of XML databases and, in particular, the challenges posed by query languages for XML data, i.e. the storage, indexing and retrieval of XML data.

18 November 2003

Ing. Sven Ubik, CESNET, Prague

High-Volume Data Transfers in Large-Scale High-Speed Networks

Abstract: This lecture will focus on both implementation and fundamental issues relating to the transmission of large volumes of data in wide-area high-speed networks, or so-called end-to-end performance. The capacity of the links forming the backbone of the Internet, both at international level and within individual national networks, currently reaches the order of gigabits per second. Links with a capacity of 10 Gb/s are already in standard operation. High transmission capacities are required for the transfer of large volumes of data in certain research projects, for example in the field of physics. End-user workstations are now commonly equipped with Gigabit Ethernet network cards, and 10-Gigabit Ethernet cards are also available. However, such high speeds, combined with signal propagation delays in wide-area networks, present certain fundamental and implementation challenges when transferring large volumes of data. The resulting throughput depends on the behaviour and interaction of all components of network communication – the network itself, the operating systems of the end-user workstations, network adaptors and applications.

25 November 2003

Dr Radim Sara CSc, Centre for Machine Vision, Faculty of Electrical Engineering, Czech Technical University in Prague

Stereoscopic Matching as a Correspondence Game

Abstract: This talk presents a game-theoretic perspective on the standard two-camera dense stereoscopic matching problem. It is argued that the nature of the problem, given general types of occlusion, requires a matching method that minimises the role of a prior model as much as possible, so that it serves no more than a regularising function. The newly formulated task leads to the Strong Sub-Kernel Problem in superoriented graphs. Strong sub-kernels are generalisations of well-known graph kernels. They possess several properties that are advantageous from a computational point of view. The definition of a Strong Sub-Kernel is provided, and the latest results concerning its existence, uniqueness and polynomial solvability are briefly discussed. A fast algorithm is outlined. Stereoscopic matching results for complex outdoor scenes as well as for standard test data are presented and discussed. The relationship to other problems in computer vision is briefly touched upon.

2 December 2003

Prof. Zdenek Strakos DrSc, UI, CAV, Prague

On the analysis of stability in numerical calculations

Abstract: Modelling real-world phenomena is based on describing a practical problem using the laws of the relevant scientific discipline (physics, chemistry, biology, economics, engineering, etc.). The resulting mathematical model is usually of infinite dimension, and its solution consists of unknown functions which can very rarely be found using analytical methods. It is therefore necessary to discretise the mathematical model, i.e. to replace the infinitely-dimensional problem of finding (continuous) functions with the problem of finding a suitable approximation of the solution with a finite—though often very large—number of unknowns. Their values are usually determined by computation on computers using numerical methods and computational tools. Over the past decades, suitable procedures have been developed for the individual steps – modelling, discretisation and computation – for many important practical problems. Both their underlying theory and the algorithms for their implementation on modern computing hardware have been thoroughly developed. However, the wide range of scientific disciplines associated with modelling and the need to understand individual issues in great depth have also led to a certain degree of separation between the individual steps. The importance of the interrelationships between the individual steps is clearly illustrated by the issue of errors. Each step is subject to an error (modelling error – discretisation error – numerical calculation error), the magnitude of which must be assessed within the overall context. In this paper, we shall focus on numerical calculations and describe some approaches to analysing and evaluating their stability. Using examples, we will demonstrate how a simply formulated yet profound understanding of technically very complex issues regarding the propagation of rounding errors in calculations can be achieved. The paper concentrates on explaining the main ideas and conclusions with a minimum of technical detail.

9 December 2003

Prof. Hiroshi Imai, University of Tokyo

Computational Analyses of Quantum Information

Abstract: Quantum information science offers new insights into computer science in various ways. Numerous theoretical and experimental studies have been conducted to establish new quantum information technology, which is believed to be far more powerful than current information processing technology. However, quantum information is often difficult to analyse compared with classical information theory, and there are still many theoretical problems to be resolved. To address these issues, we may make full use of current computing power. In this talk, we will discuss some computational analyses of problems in quantum information, some of which are followed by theoretical analysis, including (1) decoherence analyses of a quantum counting algorithm, (2) quantum channel capacity, and (3) quantum error-correcting codes. An overview of our ERATO project, which focuses on quantum computation and information, will be presented.

16 December 2003

PhD Programme Open Day

TBA