Czech 
Slovak 
Publications
| Authors | Title |
|---|---|
| P. Černo, F. Mráz | Clearing Restarting Automata |
| Fundamenta Informaticae, 2010, Vol. 104, No. 1, 17-54, DOI 10.3233/FI-2010-334 | |
| Abstract: Restarting automata were introduced as a model for analysis by reduction, which is a linguistically motivated method for checking correctness of a sentence. We propose a new restricted version of restarting automata called clearing restarting automata with a very simple definition but simultaneously with interesting properties with respect to their possible applications. The new model can be learned very efficiently from positive examples and its stronger version can be used to learn effectively a large class of languages. We relate the class of languages recognized by clearing restarting automata to the Chomsky hierarchy. | |
| Links: BibTeX, Fundamenta Informaticae | |
| R. Křižka, I. Mrázová | Kohonenovy mapy a jejich využití při analýze příčin nehodovosti v silniční dopravě |
| In J. Kelemen, V. Kvasnička, editors: Proceedings of KUZ X (Kognice a umělý život X), Slezská univerzita v Opavě (May 2010), 199-202. | |
| Abstract: Currently, road transport represents a popular, widely used kind of transportation in the Czech Republic. However, the growing number of road vehicles impacts more conflicts and car accidents leading to severe damages or even deaths. In this paper, Kohonen self-organizing maps and their variants with a dynamically adjusted topology are used to identify characteristic types of traffic accidents in Ostrava, the third biggest city in the Czech Republic. By analyzing the structure of the formed neural networks we were able to find also their main cause. | |
| Links: BibTeX | |
| I. Mrázová | Knowledge Extraction with Neural Network |
| VDM Verlag, Saarbruecken, Germany, October 2010, 165 pages. | |
|
Abstract: Recent turbulence of global economy impose strong requirements on efficiency, flexibility, and reliability
of new products and services. Unfortunately, for most techniques applicable to the design of complex adaptive systems –
artificial neural networks, fuzzy logic, cluster analysis and others – it is still complicated to interpret what they are
actually doing – in particular when processing huge sets of high-dimensional data. Yet understanding and correct interpretation
of the knowledge extracted by the applied model represent decisive issues for the ability to detect significant, e.g. novel
input patterns, to identify their characteristic features and to assess their future development. This book provides new means to handle these problems with artificial neural networks of the Back-Propagation type. Two case studies involving image classification and analysis of economical data provided by the World Bank should help shed some light on this new and exciting area, and could be useful to professionals in the field of data mining, image processing and adaptive systems, or anyone else who may be considering applying neural networks for knowledge extraction e.g. in marketing. |
|
| Links: BibTeX, MoreBooks! publishing | |
| I. Mrázová, M. Kukačka | Image Classification with GHNN-Networks |
| Proceedings of the ICMV 2010 3rd International Conference on Machine Vision (Hong Kong), IEEE, New York, 2010, 223-227. | |
| Abstract: Future multi-media technologies are expected to support efficient on-line processing of huge amounts of high-dimensional data without any special pre-processing. In this paper, we introduce a new model of the so-called Growing Hierarchical Neural Networks (GHNN) applicable to image classification without requiring advanced domain-specific feature extraction techniques. Dynamic data-dependent adjustment of both the number and position of the neurons is, moreover, supposed to improve generalization. Experimental results obtained so far for two case studies on face and hand-written digit recognition show that local features detected automatically by the network impact a transparent and compact representation of the extracted knowledge. | |
| Links: BibTeX | |
| I. Mrázová, O. Sýkora | When Is Phonetic Search with FPGAs Efficient? |
| In A. Pulka, T. Golonek, editors: Proceedings of ICSES'10 (International Conference on Signals and Electronic Systems), Gliwice, Poland (September 2010), 359-362. | |
| Abstract: Phonetic search represents a new area in information retrieval. Its goal is to search texts for all words that have the same pronunciation as the word heard and written by the user. The user is assumed to be a foreigner who uses in general a different alphabet and different transcription rules. With rapid advances in programmable hardware (FPGA), a natural idea would be to use FPGA-implementations for phonetic search. This paper provides an original methodology to assess their benefits based on the given hardware, search string, language, availability of a dictionary and length of the searched text. Supporting experiments involved German and Arabic. | |
| Links: BibTeX | |
| F. Otto, M. Plátek, F. Mráz | On lexicalized well-behaved restarting automata that are monotone |
| Proceedings of the DLT 2010 14th International Conference on Developments in Language Theory (London, Ontario, Canada), Springer, Berlin, 2010, LNCS, Vol. 6224, 352-363. | |
| Abstract: We introduce lexicalized well-behaved restarting automata as a model of the gradual lexicalized syntactic disambiguation of natural languages. This model presents a non-correctness preserving counter-part to the (correctness preserving) models of analysis by reduction of natural languages. We study two types of gradual relaxations of the correctness preserving property for monotone automata of this type. They lead to two infinite hierarchies of language classes. The basic levels of these hierarchies coincide with the class LRR of left-to-right regular languages, and the hierarchies exhaust the class of context-free languages. | |
| Links: BibTeX, SpringerLink | |
| M. Plátek, F. Mráz, M. Lopatková | Restarting automata with structured output and functional generative description |
| In Dediu, A.-H., Fernau, H., and Martín-Vide, C., editors: LATA 2010. Vol. 6031 of LNCS, Berlin, Springer (2010) 500-511. | |
| Abstract: Restarting automata were introduced for modeling linguistically motivated analysis by reduction. In this paper we enhance these automata with a structured output in the form of a tree. Enhanced restarting automata can serve as a formal framework for the Functional Generative Description. In this framework, a natural language is described at four layers. Working simultaneously with all these layers, tectogrammatical dependency structures that represent the meaning of the sentence are derived. | |
| Links: BibTeX, SpringerLink | |
| M. Plátek, F. Mráz, M. Lopatková | Towards a formal model of natural language description based on restarting automata with parallel DR-structures |
| Accepted to ITAT 2010 (2010). | |
| Abstract: We provide a formal model of a stratificational dependency approach to natural language description. This formal model is motivated by an elementary method of analysis by reduction, which serves for describing correct sentence analysis. The model is based on enhanced restarting automata that assign a set of parallel dependency structures to every reduction of an input sentence. These structures capture the correspondence of dependency trees on different layers of linguistic description, namely layer of surface syntax and layer of language meaning. The novelty of this contribution consists in (i) the extension of enhanced restarting automata in order to produce tree structures with several interlinked layers and (ii) the application of these automata to the stratificational description of a natural language. | |
| Links: BibTeX, Fulltext | |
| M. Plátek, F. Mráz, M. Lopatková | (In)dependencies in functional generative description by restarting automata |
| In Bordinh, H., Freund, R., Hinze, T., Holzer, M., Kutrib, M., and Otto, F., editors: Second Workshop on Non-Classical Models for Automata and Applications (NCMA 2010). Vol. 263 of books@ocg.at, Österreichisches Computer Gesellschaft (2010) 155-170. | |
| Abstract: We provide a formal model of a stratificational dependency approach to natural language description. This formal model is motivated by an elementary method of analysis by reduction, which serves for describing correct sentence analysis. The model is based on enhanced restarting automata that assign a set of parallel dependency structures to every reduction of an input sentence. These structures capture the correspondence of dependency trees on different layers of linguistic description, namely the layer of surface syntax and the layer of language meaning. The novelty of this contribution consists in the formal extension of restarting automata in order to produce tree structures with several interlinked layers and in the application of these automata to the stratificational description of a natural language. | |
| Links: BibTeX, Fulltext | |
| D. Průša, J. Stria | System for On-line Recognition of Handwritten Mathematical Formulae |
| Research Report CTU-CMP-2010-19, Center for Machine Perception, K13133 FEE Czech Technical University, Prague, Czech Republic, December 2010 | |
| Abstract: We present a system for on-line mathematical formulae recognition. This is an incremental work. We describe our method and summarize issues we have encountered in the previous implementations. Based on the experience, we propose new ideas how to improve robustness and accuracy. We also report our intension to develop a web application that will make the system publicly available. | |
| Links: BibTeX, Fulltext | |