Education, PhD studies and seminars

Institute of Computer Science AS CR is a public research organization, which, according to its founding charter does not have a doctoral degree program. The Institute focuses on basic research, however, employs PhD students and post-docs in the areas covered by the research institute. PhD students and postdocs with a serious interest may send CV to

Theoretical Computer Science

  • Abstract algebraic logic
  • Complexity measures in neural networks
  • Substructural, fuzzy logics and residuated lattices

Nonlinear Modelling

  • Satellite data for environmental modeling (diploma thesis)
  • Methods of data reduction in environmental systems
  • Weather and air quality modeling and forecasting - Mathematical and statistical methods for data assimilation in complex dynamical systems
  • Modern semiparametric methods and their application to modeling and evaluation of meteorological and environmental fields

Medical Informatics and Biostatistics

  • Survival analysis of multivariate censored data
    (Valenta, First Faculty of Medicine, Charles University in Prague, Biomedical informatics)
  • Robust image analysis in the evaluation of molecular genetic studies
    (Kalina, CTU or MFF – by agreement)
  • Statistical methods for dimensionality reduction and classification for multivariate data with applications in biomedicine.
    (Valenta, First Faculty of Medicine, Charles University in Prague, Biomedical informatics)

Optimization and Systems

Fundamental Topics

  • Comparisons of abilities of shallow and deep networks (more information)
    Recently, feedforward networks with several hidden layers were successfully applied to various pattern recognition tasks. These experimental results pose theoretical questions concerning comparisons of abilities of deep and shallow feedforward networks with various types of computational units. The goal of the thesis is investigation of the role of the number of hidden layers from point of view of the overall number of computational units, performance on high- dimensional tasks, generalization, and data-mining.
  • Boolean nonlinear factor analysis (more information)
    The aim of the dissertation is to create effective methods for nonlinear Boolean factor analysis of signals of different properties using the paradigm of neural networks. Work will build on previous research of supervisor and his collaborators. In the work will be defined model of data suitable for this task and and its base new masure which allows a comparison with existing methods that deal with similar problems: eg. special clustering methods, nonnegativní matrix decomposition, another neural network based methods and classical statistical methods. Research is one of the current issues of methods for extraction of information from large dynamic digital data including internet content.
  • Brain computer interface (more information)
    The aim of the dissertation is to contribute to the development of the up to day research of "Brain Machine Interface" (Brain Machine Interface, or Brain Computer Interface), ie. To the transformation of the physical manifestations of brain thought to action, or conversely, the signals from the sensors to the perception. At this stage of the research dissertation will focus mainly on detection signals to locate / improving some signal analysis so as to achieve greater recognition accuracy as well to be posssible detection of multiple signals, all in the shortest possible time. Research is one of the highly topical issue of contactless interaction man machine. The resulting product could e.g. immediately find application in the field of rehabilitation of some neurological diseases.
  • Research and comparison of methods of group factor analysis for fMRI data (more information)
    The aim of investigation is development of methods for fMRI data analysis as a tool for Brain Computer Interface research (BCI). Interest to this theme is quickly growing. Problem of such data analysis is huge size of data arrays. One of the data dimensions corresponds to number of voxels and exceeds hundreds of thousands. So this types of data overlays some restrictions on methods which can be used. Inceptor should focus on analysis, testing and modification of existing methods as well as creating of new methods. One of approaches suitable for such task is factor analysis. However standard methods of factor analysis were developed for two-dimensional data only. Recently the parallel factor analysis (PARAFAC), tensor ICA and others methods were developed for three-dimensional data. First and second dimensions are usually given by number of time slices and by number of voxels in fMRI. Third dimension depends on application. If one use number of subjects as third dimension, we have method of group factor analysis, and the result is common for all subjects. To verify efficiency of methods developed, one should apply them to fMRI data obtained before and after subject training by BCI based on motor imagery. Comparing result given by different methods one should make conclusions about their efficiency for fMRI data analysis. (Ing. Dušan Húsek, Ph.D., trainer-specialist: Prof. Ing. Alexander Frolov, DRSc.)