Machine-translated page for increased accessibility for English questioners.
Colloquium programme with abstracts for the Spring 2014 semester
- 25 February 2014
- Prof. Dr Doğan Kesdogan, Faculty of Business, Economics and Management Information Systems, University of Regensburg
- Interdependent Security and Compliance in Service Selection
- Abstract:
Service-Oriented Computing has gained significant importance in recent years
in both research and industry. Concepts such as cloud
computing or Everything-as-a-Service (XaaS) are the latest trends and are
used extensively to implement business processes. However, the
cross-organisational and cross-border implementation of service-based
processes presents challenges regarding security and privacy.
These properties can be regarded as part of the non-functional quality
aspects of services, i.e. non-functional Quality of Service (QoS) properties.
Current approaches to service selection treat security either as a single QoS attribute or as several mutually independent quality properties. However, this view is inadequate, as security objectives do not exist in isolation but are interdependent. Formal methods, on the other hand, are limited to a fixed set of security objectives and do not account for the interdependence of objectives either.
In this talk, we present an approach that assesses the quality of service compositions with regard to interdependent security objectives whilst taking compliance requirements into account. Our work utilises the concept of structural decomposition, which estimates the impact of individual quality attributes on a security objective. Individual security objectives are expressed as utility functions that enable the assessment of the fulfilment of a security objective. This, in turn, enables the definition of domain models for any set of security objectives and the modelling of interdependence by means of functional interdependence. The result is a Multi-Objective Optimisation (MOO) problem. We present initial evaluation results of transforming domain models into MOO problems and solving them using state-of-the-art genetic algorithms. - 4 March 2014
- Mgr. Tomáš Jirotka, GoodData
- Applications of Algebra in Data Analytics
- Abstract: GoodData provides its customers with comprehensive management of their data, including tools for its analysis. Data stored in relational databases are described by an abstract logical data model, the description of which utilises the properties of algebraic structures. Thanks to these properties, it is possible to validate users’ analytical queries and optimise them. The lecture will explain the principles of the abstract mathematical model and the implications of its properties for the average user – the analyst.
- 11 March 2014
- Dr Luděk Smolík, Siegen, Germany
- Is there only one correct way to understand quantum cryptography?
- Abstract: Quantum cryptography is one of the few practical applications of quantum mechanics and specifically utilises the measurement of individual quantum states of elementary particles as carriers of quantum information. Quantum mechanics itself is a theoretical description of the behaviour of elementary particles and, to date, has brilliantly predicted the results of even the most subtle experiments in the microcosm. It should be noted that the concept of ‘predicting a result’ in quantum mechanics differs fundamentally from that in other classical physical theories. Quantum mechanics provides only the probability of the measured value occurring. A generation of physicists during the first two-thirds of the last century worked intensively to explain these oddities and the lack of clarity in quantum mechanics. Today’s approach to these unresolved paradoxes, by contrast, is almost pragmatic. The approach of quantum cryptography is also predominantly optimistic and pragmatic, as it bases its security precisely on these unexplained—and perhaps even inexplicable—quantum oddities. This paper offers a brief discussion of the history of interpretations of quantum mechanics, the concepts of reality and locality, the Einstein–Podolsky–Rosen thought experiment, and the implications for quantum cryptography.
- 18 March 2014
- RNDr. Igor Peterlík, Ph.D., Faculty of Information Technology, Masaryk University
- Interactive computer-based surgical simulations and modelling
- Abstract:
Over the last decade, interactive virtual models of living tissues
have been developed and used primarily in the field of medical education
and training. Advances in hardware performance, human-computer
interaction and numerical modelling have opened up new horizons:
computer simulation is now being directly integrated into clinical
procedures such as pre-operative planning and intra-operative
guidance. Generic models previously used in the context of medical
training are being replaced by specific ones, tailored to the actual
patient and the intended procedure. New interfaces are being developed
to facilitate interaction between the surgeon and the computer.
Furthermore, much higher standards are required regarding the robustness
and stability of the underlying numerical algorithms.
In this talk, we provide a general overview of the state-of-the-art methods used in soft-tissue simulations, focusing mainly but not exclusively on patient-specific modelling. We demonstrate that the field of medical simulations is an exciting area of research where techniques, methods and algorithms developed within the fields of informatics and computer science are utilised to improve the quality of healthcare and reduce risks for both patients and doctors. We also present our scientific contribution to this field, focusing on needle insertion planning and guidance in interventional radiology and the prediction of tumour position during laparoscopic liver surgery, based on biomechanical simulation and augmented reality.
Since graduating, Dr Peterlík has spent several years as a postdoctoral fellow and researcher at the following institutes:
- 2009–2011: INRIA, Lille, France
- 2011–2012: University of British Columbia, Vancouver, Canada
- 2012–2013: Institut Hospitalo-Universitaire, Strasbourg, France
- 25 March 2014
- Ing. Jiří Jaroš, Ph.D., FIT VUT
- Supercomputer simulations of ultrasound propagation in the human body
- Abstract:
Realistic simulations of the propagation of ultrasound waves in living tissues offer a wide range of applications. One very promising area is high-intensity focused ultrasound, used in non-invasive surgery to treat cancer or to stop internal bleeding. This method works on the principle of emitting a beam of concentrated ultrasound waves into the tissue. The acoustic energy reaches such a high level at the focal point that it causes cell necrosis, whilst leaving the tissue outside the focal point undamaged.
For successful and gentle treatment, it is essential to position the focal point of the ultrasound waves with precision. However, precise targeting is very difficult to achieve due to the distortion of the ultrasound waves caused by their passage through tissues with different properties (layers of fat, soft tissue, cartilage and bone). This is where computer simulations come into play, providing precise surgical plans for individual patients. Realistic simulations, however, require enormous data sets and computational power that only supercomputer systems are capable of providing.
One of the modern acoustic simulation systems is the k-Wave toolbox (http://www.k-wave.org/), which was developed as a Matlab extension in 2009 at University College London. In 2011, development was transferred to the Australian National University, where I also joined the project. Within two years, we succeeded in creating a high-performance implementation in C++ capable of utilising up to 4,096 processor cores efficiently. This implementation made it possible to increase the size of the simulation domain by three orders of magnitude, whilst maintaining realistic simulation runtime. In 2013, development of the toolbox was transferred back to Europe. Development of the physical model is taking place in London, whilst implementation, debugging and scaling are being carried out at FIT, Brno University of Technology, with the support of supercomputer allocations under the IT4I and PRACE projects.
The lecture will outline the steps and challenges that led from the development of the first sequential version to a highly efficient parallel variant, which has been tested and is now running in production on the world’s leading supercomputers (Raijin, HECToR, Legion, Emerald, Anselm). Furthermore, recent advances in the use of hybrid systems (GPUs) and new decomposition techniques, which should enable scaling beyond 10,000 cores, will be presented. Finally, I shall briefly summarise my observations and experiences from my 2.5 years at ANU Canberra and UCL London.
- 1 April 2014
- RNDr. Petr Holub, Ph.D., Faculty of Computer Science, Masaryk University
- Image Compression on Massively Parallel Architectures
- Abstract: Advances and the practical availability of massively parallel computing systems, such as graphics processors and many-core commodity accelerators, have dramatically changed the landscape of high-performance computing. This talk will discuss the applicability of such systems to the field of image and video compression. I will discuss various types of algorithms used specifically in image compression, such as data transformation, motion estimation, or entropy coding, and their suitability for massively parallel architectures. As an active contributor to research in this field, I will illustrate these issues using our research findings, which have pushed the boundaries of the applicability of widely used and established algorithms on such systems.
- 8 April 2014
- Prof. RNDr. Jaroslav Nešetřil, DrSc., Dr.h.c.mult., Faculty of Mathematics and Physics, Charles University
- Sparse and Dense Classes of Structures with Algorithmic Implications
- Abstract: In this lecture, we shall present the dichotomy between sparse and dense structures, as defined by the author together with P. Osson de Mendez, in many (sometimes surprising) forms. It appears that this distinction provides a useful framework for problems that can be solved efficiently using algorithms.
- 15 April 2014
- Assoc. Prof. PaedDr. RNDr. Stanislav Katina, Ph.D., Faculty of Science, Masaryk University
- (Semi)landmark-based shape analysis: Landmarks to curves to surfaces
- Abstract: A three-dimensional image of the human face can be captured by, for example, stereo-photogrammetry. Subsequent smoothing is often required due to imperfections in the surface representation. Compared with anatomical landmarks, curves have the advantage of providing a much richer description of facial shape. In this talk, this is explored in the context of identifying certain anatomical curves (valleys or ridges) and geodesics between two carefully chosen anatomical landmarks. These curves, defined by semi-landmarks, are automatically identified by curvature in particular local surface patches, the detection of slope discontinuities in local principal curves or optimised surface cuts. A complete standardised surface representation is then obtained by interpolation across the relatively flat surface patches between the identified curves. Each level of this hierarchical structure (landmarks, curves, surfaces) lends itself to shape analysis. For the purposes of visualisation, a high-resolution template can be fitted to the semi-landmark surface by warping. Where the semi-landmarks and/or template have a symmetric structure, this leads to natural ways of quantifying asymmetry. These methods will be illustrated using facial images collected in various studies.
- 22 April 2014
- RNDr. Jana Tůmová, Ph.D., Faculty of Informatics, Masaryk University
- Intersection of Theoretical Computer Science, Robotics, and Automatic Control
- Abstract: In this talk, we provide an overview of a research area that has recently attracted considerable attention: the application of formal verification methods from computer science to the automatic control of reliable robotic systems. We present the approach of ‘correct-by-design’ control of complex dynamical systems, particularly mobile robots and autonomous vehicles, whilst discussing the benefits and challenges associated with bridging control and computer science techniques. We summarise the recent contributions to this field of research and outline the currently open research questions. The talk draws on experience gained during collaborations with Boston University, the Massachusetts Institute of Technology and the Royal Institute of Technology on projects relating to robot motion planning, autonomous car development and multi-robot control, respectively.
- 29 April 2014
- Alexandru Popa, Ph.D., Faculty of Informatics, Masaryk University
- The Power of Approximation Algorithms
- Abstract:
Under the widely held conjecture that P is not NP, no polynomial
time exact algorithms exist for many important problems. However,
suboptimal solutions (which run in polynomial time) are, in most cases,
sufficient for practical applications. Moreover, approximation
algorithms allow us to explore the structure of NP-hard problems and
distinguish between different levels of difficulty that these problems
exhibit in theory and practice.
In this talk, I shall attempt to provide an overview of the most famous results in the field of approximation algorithms. Interestingly, many problems which appear unrelated at first glance can be tackled using similar techniques. I shall therefore also survey the main methods used to date to derive approximation algorithms and to prove the hardness of approximation.
In the final section, I will present some of my recent work. I will present some novel results concerning NP-hard problems with applications in areas such as wireless networks, computational biology and pattern matching. Finally, I will discuss an intriguing problem which was previously believed to be NP-hard, but which turned out to be solvable in polynomial time.
- 6 May 2014
- Prof. Ing. Josef Psutka, CSc., Centre for Computational Linguistics, University of West Bohemia in Plzeň
- Eliminating language barriers for disabled viewers of Czech Television
- Abstract: The lecture will focus on the preliminary results of the project “Elimination of language barriers for disabled viewers of Czech Television”, which is being carried out at the Faculty of Applied Sciences, University of West Bohemia in Plzeň, with the support of TAČR, on behalf of Czech Television. The project has four objectives: a) subtitling of ‘live’ TV programmes; b) creation of an accompanying audio track for TV broadcasts; c) automatic generation of sign language; d) assisted cataloguing of TV programmes. The lecture will present the latest results of the project, particularly in the area of subtitling ‘live’ TV programmes. The solution is based either on the direct processing of the accompanying audio track or on the use of a so-called ‘shadow speaker’. Both methods of subtitling are already in use today and are applied in practice to subtitle programmes on Czech Television in selected (and so far processed) genres.
- 13 May 2014
- Mgr. Pavel Zahradníček, Ph.D., CzechGlobe, Global Change Research Centre of the Czech Academy of Sciences
- The use and processing of measured and predicted time series of meteorological variables
- Abstract: The aim of the lecture is to familiarise the audience with meteorological data and how to work with it: how measurements are actually carried out, what errors time series of meteorological variables contain, and how to use mathematical and statistical methods to remove non-climatic factors from these time series. The first part of the lecture will therefore be devoted to data quality control, homogenisation and the imputation of missing values, including the possibility of creating both regular and irregular grids using interpolation methods. The next part of the lecture will deal with climate modelling and the potential use of supercomputers. The results of climate modelling up to the year 2100 for the territory of the Czech Republic will be presented. An interesting aspect of meteorology is the relatively new field of energy meteorology, in which, based on numerical forecast models, the output of solar and wind power stations is predicted at hourly intervals, which is important for stabilising the electricity grid.