Programme of colloquia with abstracts for the Autumn 2000 term

3 October 2000

Mgr. Michal Kozubek, Dr., Faculty of Informatics, Masaryk University, Brno

Automated acquisition and analysis of biomedical images using computer-driven optical microscopy

Abstract: We have recently described a high-resolution cytometry (HRCM) technique (Cytometry 36:279–293) which enables the automated measurement of large numbers of cells and the collection of statistically significant data for the investigation of the human genome.
The HRCM system comprises a fully motorised fluorescence microscope and a cooled digital CCD camera, both fully controlled by a high-performance computer which performs both image acquisition and related on-line image analysis. It is capable of analysing microscope slides containing cell nuclei in both 2-D and 3-D. Using HRCM, it is possible to analyse multi-colour preparations, including those containing UV-excited dyes, as well as repeatedly stained preparations by re-acquiring individual cells.
Very recently, we have built a second HRCM instrument which enables switching between conventional and confocal modes. The confocal mode is achieved using a 5% Nipkow disc optimised for fluorescence imaging. It offers better resolution but at a lower speed. The user can select which fluorochromes are imaged in confocal mode. Consequently, the system can be configured for a variety of imaging modes, ranging from rapid 2-D analysis to high-resolution confocal 3-D studies.
Both instruments are controlled using our own software. The software has been optimised for speed and reliability. Special algorithms have been developed for the accurate segmentation of cell nuclei, as well as genes or chromosome domains. Computers are also being used to correct aberrations present in optical microscopy and to determine optimal imaging conditions. A special effort is being made to develop a suitable network architecture for biomedical image processing using the facilities of the Supercomputing Centre at Masaryk University.
Both hardware and software aspects will be discussed, as well as the future goals of this highly interdisciplinary research.

10 October 2000

Prof. Ing. Vaclav Skala, CSc., University of West Bohemia, Plzeň

Algorithm Complexity and Data Visualisation in a Parallel Environment

Abstract: Computer graphics and data visualisation are characterised not only by the complexity of the algorithms themselves, but above all by the volume of data being processed and the requirement for interactivity. It is therefore necessary, in addition to the ‘classical’ approach, to consider the possibilities of pre-processing, distributed and parallel processing.
The processing of volumetric data (CT, MRI and PET data) is a case where computational complexity increases non-linearly with the resolution of the input data. The selection of isosurfaces is a common task, yet one that leads to a large volume of geometric data, whilst requiring a high degree of interactivity. The lecture will present results, in particular, relating to the accuracy of isosurface generation and the reduction of surface meshes using parallel computing on the Windows NT platform.

17 October 2000

Prof. Eva Hajicova, DrSc., Faculty of Mathematics and Physics, Charles University, Prague

‘Hot’ topics in computational linguistics (this year’s world conferences through the eyes of a Prague linguist)

Abstract: This lecture will summarise and evaluate the findings from two world conferences on computational linguistics held this year (COLING in Saarbrücken and the annual international conference of the Association for Computational Linguistics in Hong Kong) and will outline the directions that appear to be the most topical and promising for the near future of this discipline, both in the field of theoretical research and in its applications.

24 October 2000

Prof. RNDr. Jiří Zlatuska, CSc., Masaryk University, Brno

Information Technology and Economic Growth

31 October 2000

Stefan Schwoon, Technical University of Munich

Verification Problems for Pushdown Systems

Abstract: Pushdown systems are a class of infinite-state systems within the hierarchy of Process Rewrite Systems introduced by Mayr and Esparza. They can be viewed as pushdown automata used not as language acceptors but as transition systems. The interest in this class stems from the fact that they can be viewed as a model of sequential programmes with recursive procedures. In particular, efficient algorithms have been developed for them in the field of dataflow analysis and for model-checking linear-time logics. In this talk, I will present results for the latter and some examples. Most of the material is based on a paper published at CAV 2000, with the exception of some recent extensions.

7 November 2000

Doc. Miloslav Dusek, CSc., Palacký University, Olomouc

Quantum cryptography

Abstract: Quantum cryptography is a new method for secure communication. Whilst the vast majority of classical cryptosystems currently in use rely on the one-way complexity of certain mathematical operations, the security of quantum cryptography is guaranteed by the fundamental laws of quantum physics. The key point is that non-orthogonal states of quantum particles are used for the transmission of a cryptographic key. Quantum cryptography can be effectively realised by means of quantum optics. In this case, the ‘alphabet’ is represented by the quantum states of photons, and optical fibre serves as the transmission channel.

14 November 2000

Doc. Ludek Matyska, CSc., Faculty of Informatics, Masaryk University, Brno

GRID — Call for Papers in Computer Science

Abstract: The distributed nature of computer networks, and in particular the Internet, on the one hand, and the increasing computational demands of truly realistic computational experiments on the other, have led to the emergence of GRID – techniques, methods and environments for large-scale distributed computing. The creation of a fully-fledged GRID currently represents one of the greatest technical challenges ever faced by computer science. It is necessary to find answers to a whole range of fundamental theoretical questions, to master ‘large-scale’ programming, and at the same time to ensure a high level of security for the environment created.
The lecture will focus on selected issues within the field in which the MetaCentra group is engaged:
- measurement and analysis of GRID properties
- resource planning and alignment with application requirements
- GRID security

The lecture will also briefly introduce what is currently the largest European project in this field – DATAGRID – and will present experiences from an extensive experiment in which the MetaCentra group participated.

21 November 2000

Prof. Jaroslav Pokorny, CSc., Charles University, Prague

XML and Databases

Abstract: With the advent of new ways of publishing information sources, the problem of how to integrate heterogeneous data has emerged in a new form. Alongside structured data from relational databases, plain text, images, web pages and the like are also available. Using the XML language, semi-structured data can be derived from this unstructured data; this data carries information about its type either internally (via tags) and/or via regular expressions in a specific description resembling a database schema. Structured data from databases can be described in the same way. The corresponding data model is expressed using a tree, or a more general graph, or even, for example, a set of functions. The aim of the lecture is to present both XML data models and query languages that enable queries to be formulated over an XML database. The results of our own research will also be presented – a functional model of XML data and the possibilities for using typed lambda calculus for querying.

28 November 2000

Prof. Paolo Ciaccia, University of Bologna, Italy

Indexing metric spaces with M-trees: basic principles and recent advances

Abstract: Metric trees represent a major recent achievement in database indexing technology for the management of similarity queries (range and k-nearest neighbours). What distinguishes metric trees from multi-dimensional (or ‘spatial’) indexes is that they operate on arbitrary metric (rather than vector) spaces, which is particularly relevant for querying multimedia repositories and for supporting data mining activities. The first part of the seminar will illustrate the basic principles of the M-tree, which is indeed the state-of-the-art metric index today. The second part will discuss some recent issues related to similarity query processing, including the use of user-defined and approximate distance functions and the management of complex queries across multiple metric spaces.

5 December 2000

Doc. Ivan Kopecek, CSc., Faculty of Informatics, Masaryk University, Brno

Computers and Playing Poker (Dialogue Automata)

Abstract: The dialogue model based on dialogue automata relates the internal states of a dialogue participant to the attributes of dialogue utterances by means of transition and output functions. Consequently, this model relates emotions (described by attributes of the internal states) to prosody (described by attributes of the dialogue utterances) and other quantities that are fed into the model. This approach can be used to predict the user’s behaviour as well as to determine the value of unknown attributes within the context of a dialogue, and it raises some new and interesting theoretical perspectives and problems.

12 December 2000

Prof. Hynek Hermansky, Oregon Institute of Technology, USA

Current Methods for Acoustic Analysis of Speech

Abstract: The basic principles of automatic speech recognition will be reviewed. The concept of short-term analysis and spectrograms will be introduced, together with the linear model of speech production. The rationale for spectral envelope estimation in ASR, as well as basic techniques for its estimation—such as homomorphic analysis and linear predictive analysis—will be presented. The cepstrum, as an approximation to the Karhunen–Löve transformation, and cepstral lifters, as means of modifying the properties of simple Euclidean cepstral distances, are also introduced here.
The concept and basic properties of the modulation spectrum of speech are explained, and its historical use in predicting speech intelligibility in auditoriums is mentioned.
RASTA filtering is introduced as an extension of FIR filtering used in dynamic feature estimation, and the reasons for its robustness to changes in communication environments are explained.
The need for data-driven feature extraction is discussed, and techniques for designing discriminant spectral bases (as alternatives to conventional cepstral analysis) and discriminant RASTA filters are described, along with recent results from their applications in automatic speech recognition and speaker recognition.
Finally, the recently introduced method of speech recognition based on temporal patterns of spectral energies is described, and its inherent advantages in speech recognition in adverse environments are discussed.

19 December 2000

Dr Erszebet Csuhaj-Varju, Computer and Automation Research Institute, HAS, Budapest

On networks of Watson-Crick DOL systems

Abstract: Networks of Watson-Crick $D0L$ systems are distributed systems of language-determining devices inspired by Watson-Crick complementarity, a fundamental concept of DNA computing.
In this talk, we discuss the behaviour of particular variants of these networks using different communication protocols, and we describe the growth in the number of strings at the nodes whilst the system is in operation. We present results on the computational capacity of these language-generating devices and on the status of some important decidability problems, such as the existence of a so-called ‘black hole’ in the network.