Seminar program for 2004/2005

Spring 2005

24. 2. 2005

Introductory Seminar of the Spring Semester

Program:
Information on the seminar concept in the spring semester.
Agenda of the seminar.
Discussion.

3. 3. 2005

L. Bartek

Graphical user-friendly interface for dialog processing

Abstract
The dialog interface allows you to increase the accessibility of the application for other user groups. The aim of this lecture is to acquaint students with the basic principles of generating dialog interfaces from the graphical interfaces and further to specify the requirements for the graphical interface so that they can safely convert these interfaces into a dialog form.

10. 3. 2005

P. Cenek

DS Management

Abstract:
First, the dialog systems and their structure will be introduced and categorized according to the dialogue management approach. The rest of the talk will focus on the dialogue management in frame-based dialogue systems. Dialogue strategies that control the dialogue will be introduced and representation of the dialogue task / domains will be discussed. The talk will be completed with a brief overview of my work.

17. 3. 2005

M. Dokulil, I. Kopeček, J. Thomas

Bach, Goedel and Informatics Dedicated to the 320th anniversary of Bach's birth and to the 70th anniversary of Goedel's papers on the consistency of the axiom of choice.

Abstract:
What is it about Bach's music that so fascinates and enchants so many mathematicians and computer scientists? What are the parallels between Bach's music and Goedel's work, and between baroque music and today's informtics? There are many such questions, which are both controversial and provocative, while at the same time being inspiring and fascinating. This presentation will touch on some of these points and try to illustrate them, in the languages ​​of music as well as in informatics.

24. 3. 2005

D. Novák

chiDistance: A Distributed Data Structure for Similarity Search

Abstract
The need for search mechanisms based on data content rather than data attributes has recently led to the formation of metric-based similarity retrieval. The computational complexity of such retrieval and the large volume of processed data call for distributed processing. In this presentation, we suggest "chiDistance" and the distributed data structure for similarity search in metric spaces. The structure is based on the idea of ​​a vector-based iDistance index method which transforms the issue of similarity search into a one-dimensional range search problem. A Peer-to-Peer system based on the Chord protocol is created to distribute the storage space and parallelize the execution of similarity queries. In the experiments conducted on our prototype implementation we study the system performance focusing on several aspects of the parallelism of the range search algorithm.

31. 3. 2005

J. Kohout

Computer Algebra Systems in modeling biologic-medical processes

Abstract:
Computer Algebra Systems (CAS) can be defined shortly as computation with symbols representing mathematical objects - numbers, functions, systems of equations, groups, rings, etc. The adjective "algebraic" emphasizes that in many cases the goal is to express the answer in a closed formula or symbolic expression instead of using floating-point approximation. CAS has a wide use in mathematical models in a diverse range of disciplines ranging from natural sciences through social sciences to economics. Generating such models is the main topic of the lecture.

7.4. 2005

J. Špojcar

TBA

14. 4. 2005

V. Ulman

TBA

L. Boháč

TBA

21.4. 2005

M. Křipač

TBA

28. 4. 2005

P. Drasil

TBA

B. Zimmerová

TBA

5.5. 2005

I. Peterlík

TBA

R. Pospíšilová

TBA

12.5. 2005

P. Fibich

TBA

19 May 2004

P o s ter S e s s s o n

Autumn 2004

30. 9. 2004

Introductory Seminar Seminar

Information on Seminar Concept in the Autumn Semester.
Agenda of the seminar.
Discussion.

7. 10. 2004

Luděk Bartek

Generate dialog interfaces

Abstract
In this lecture, we will be concerned with generating individual types of dialogue strategies within the DIG system. Demonstrating methods for generating dialogue strategies with system initiative and dialogue streets with a mixed initiative. I will think about some of the problems that I raise and their reseniies.

14. 10. 2004

O. Krajíček

GSIP - Alternative Web Services Invocation Protocol / Infrastructure

Abstract:
Generic Service Invocation Protocol (GSIP) aims to provide an alternative communication infrastructure for applications based on Web Services technology. It is designed for environments where the traditional SOAP based approach imposes drawbacks on performance and scalability, leading to wasting resources and / or processing power. Such problems may emerge in various environments, such as large-scale information systems, Grid computing, high-performance or mobile computing applications. The GSIP is designed to address the shortcomings, by introducing a service-specific invocation model and corresponding infrastructure and by using alternative data encoding schemes. This allows for expandability and implementation of more advanced communication models. GSIP design and architecture, along with some basic performance studies will be presented.

M. Batko

Indexing Techniques in Peer-to-Peer Networks

Abstract:
Since the emergence of high-speed networks, interest in distributed systems has increased considerably. There are many research challenges in this area. However, a distributed storage system may be the most interesting one. Such a system would allow to store some general blocks of data (files, documents, etc.) and distribute them among network nodes (computers participating in a network). The data can be retrieved using queries issued by participants of the network. A well-defined paradigm for solving this need is available nowadays: a Peer-to-Peer (P2P) architecture. The presentation will provide a brief survey of current P2P indexing techniques with emphasis on P2P system with support for similarity searches.

21. 10. 2004

P. Moravec

Distributed Algorithm for LTL Model Checking

Abstract
We present a new distributed-memory algorithm for enumerating LTL modeling that is designed to run on a cluster of workstations communicating via MPI. The detection of accepting cycles is based on computing maximum accepting predecessors and the subsequent decomposition of the graph into an independent predecessor subgraphs induced by maximum accepting predecessors. The influence of ordering on the algorithm performance is discussed. Experimental implementation of the algorithm shows promising results.

4 November 2004

T. Hudík

Text Categorization with Support Vector Machine

Abstract
It will be given a short theoretical background about Support Vector Machines. This seminar will describe some of the results obtained from testing the influences of selected important parameters of SVM applied to text categorization. The main object was to verify whether the results obtained with standard, publicly available data sets (the traditional Reuters text documents and the 20Newsgroups) could be applied to real medical text documents from various Internet resources used by physicians. The research also focused on features like document similarity, balance of categories, presence of common words (stop-words), and data volume. The results of experiments have shown that there may be typical problems with setting up parameters for some real data. Especially the medical documents provided worse results because the real-data categories were not well balanced and the documents in different categories were mutually rather similar - ie, overlapping classes. As a result, SVM could not always find sufficient good separating hyperplanes as it mostly did for `trouble-free 'datasets like Reuters or 20Newsgroups.

11. 11. 2004

L. Pokluda

Haptical navigation of visually impaired persons in buildings

Abstract
The spatial orientation of the visually impaired persons has some specialties. Instead of global knowledge of the scene gained by sight, the visually impaired person's knowledge of the scene depends on the continuous addition of information about the local space to the global scene during the walk through through. Preliminary tests have demonstrated the necessity of a whole range of methods for supporting scenic walkthrough of visually impaired people. According to different individual ability of independent search scenes, the computer has to show scene information in different ways. Persons who are able to actively explore the scene need a kind of haptic model of real environment. On the other hand, some visually impaired people are not able to search the scene actively and prefer a passive let-by-hand approach. These two extremes fall on each side of the haptic navigation continuum which we want to present.

18. 11. 2004

V. Kovalčík

Accelerating rendering scene using occlusion queries

Abstract
Large virtual worlds are becoming common today. Unfortunately, computers are not powerful enough to display them using simple brute force methods. Therefore, advanced algorithms have to be used. We will present an algorithm for accelerating the rendering process using an occlusion query function found on a modern graphic card.

25. 11. 2004

J. Hubený

Image Segmentation via Level Set Methods

Abstract
One of the main goals in computer vision is to locate certain objects of interest and distinguish them from the background. This is called image segmentation. The level set methods are nowadays widely used in various fields (computer animation, CAD design, fluid dynamics, robotic navigation, image segmentation and denoising). The use of level set methods in image segmentation will be presented. Level set methods belong to the active contour models. These methods are based on the tracking of the interface motion. In the image segmentation, the interface represents the boundary between the objects of interest and the background. Starting with some initial boundary, the method moves the interface under a certain velocity field, which is derived from the image data, to the desired boundaries. A major advantage of the level set approach is its ability to segment arbitrary complex objects without any a priori knowledge of their topology and their amount. Moreover, the presented methods can be used without significant modifications in any dimension.

2 December 2004

F. Procházka

Universal Information Robot

Abstract
The project of designing a computer system called Universal Information Robot will be presented. Surprisingly, the main topic of the design process (7 years long) is simplification. We will explain the basic modules, describe the process of building such a system and briefly discuss some related areas such as modern trends in artificial intelligence, self-referencing and self-similarity (fractal geometry) and transparent intensional logic. The application of such a system is quite large. Some projects and grants, where the Universal Information Robot is used (crisis management, oncology, project management) will be presented as well.

9. 12. 2004

P. Šmerk

Rules for Morphological Disambiguation

Abstract
State-of-the-art rule-based tools for morphological disambiguation use either manually crafted rules or rules to learn from manually annotated data. We will present a new method of learning rules for morphological disambiguation using only unannotated data. Inductive logic programming and active learning are employed. The induced rules display very promising accuracy. Also, the probable limitations of the proposed method will be discussed.

16. 12. 2004

P o s ter S e s s s o n

Presentations will be created in the autumn semester