Embed Size (px)
Citation preview
00%$010!@%@%$#@##⬂!@%$@*@*#�$#$@%$#@##&10#&$%$&@!1101010011010!@%$@*@*#�$#$@%$#@##&1@1%001110011*@&01010100&#*@!1@1!0101###0101#11$#$#@#11010@%$@*@*#�*@#�#$%*IT4Innovations#&0110&$%$01@1$#@%$#@##%$##&#*@!@!1010*@$%011$#national01$%@&@1@00%$#@&#*0#$#$111$#$@%$%%$##&#@$#@#&01supercomputing&011#111$#$#&#*@@100%$010!@%@%$#@##&#*10#&$%$¢er$@0@&1@%$0111001%@10110111$##@##&#*@!@!11$%011$#@!@!10101#$%*@*111$#$#@#11010!@%$@*@*#�*@#101010011010!@%$#�$#$@%$#@##&1@1%001110011*@&01010100&#*@!1@1!0101#$%
@!0101010011010!@%$@*@*#�$#$@%$#@##&1#*@100#$%*Abstract#Book#&0110&$%$01@1$#@10101%$#@##%$##&#*@!@!1010*@&@0@00%$#@$#$@%10101$%%$##&#@$01%@101101#@
2nd Annual Conference of IT4Innovations
www.it4i.cz
Time 21 to 22 October 2013 Venue VŠB – Technical University of Ostrava, New Hall, Room No NA2, 17. listopadu 15, 708 00 Ostrava – Poruba
Earth Sciences
21 October 2013 9:00 AM - 10:30 AM Earth Sciences
Radim Briš New Development in Reliability and Risk Modeling
New findings in two recently solved research topics will be discussed during the presentation:
1. Effective methods and algorithms for decision making processes in different risk situations. The research is focused on both industrial and medical risk.
2. Uncertainty analysis for risk models.
Martin Hájek New Challenges in the Field “Transport BIG DATA Collection and Analysis”
The research in the field of collection and analysis of data, simulation and optimisation of people and goods mobility has been very fragmented and limited to particular problems. A complex solution has been missed. The reason for that was a total lack of real data characterizing all the things related to mobility in a complex way. In the transport area, except air transport and partially public transport, there is a lack of data and existing data provides just limited knowledge of real traffic development. In other words it is not known what the start and end destination of people journeys is like, the time the journey takes, why they travel and why they choose a particular mode of transport, particular journey trajectory and the time. Even the dynamics of traffic flows is not monitored in comprehensive manner and so, in consequence, there is a low efficiency of control systems on highways and in urban and suburban areas as well. Missing real data and information is partially replaced by modeling. It should be noted that the modeling is often very imprecise due to the complexity and changeability of traffic system.
One of the significant issues of the centre RODOS to achieve the goals is the use of so called time and space independent sample methods for data collection in the model computation of actual state of transport and mobility of people in a space. To the existing profile measurements and selected examinations the time-continuous and whole-area available data from sample technologies are to be added, even they work with GPS data or signalling data from GSM network. Thanks to the fusion of new and existing sources of data far more precise dynamic models are to be gained; some “rough statistics estimations” made with minimum data. The above described way enables an optimum design and implementation of innovative traffic control systems, combined toll systems, information and support systems and other advanced applications.
Aleš Ronovský Bringing Large CFD Modeling Capabilities into Ostrava Using Code_Saturne Software
The long term objective of the current research is to develop algorithms able to handle extremely complex simulations, which are known as Computational Fluid Dynamics. This type of simulations requires large computing resources and could be only performed on parallel computers and ANSELM is such a computer. To be able to solve different CFD problems we need different software. Code_Saturne is a general purpose computational fluid dynamics open source computer software package that is based on a co-located Finite Volume approach that accepts meshes with any type of cell (tetrahedral, hexahedral, prismatic, pyramidal, polyhedral...) and any type of grid structure (unstructured, block structured, hybrid, conforming or with hanging nodes...). It is very portable, built in massively parallel way and could be coupled to different software as e.g. Code_Aster. Cooperation between IT4I and developers of Code_Saturne brings us possibility to be not only user of that software, but even to participate in developing new capabilities of new versions of that code. This presentation will be about bringing more capabilities into Code_Saturne by full hybrid Mesh_Multiplication package developed by researchers from IT4I, benchmarking of ANSELM using Code_Saturne, solving some real complex cases on ANSELM and comparison of performance with some other supercomputers using same test cases and scales.
New Materials
21 October 2013 11:00 AM - 12:30 PM New Materials
Jaroslav Hamrle Development of New Materials for Magneto-Optical Applications
Magneto-optical (MO) phenomenon is unique as it provides optical non-reciprocal behaviour, i.e. the light phase or absorption is different for forward and backward directions of the light propagation. Intrinsically, it originates from broken crystal symmetry due to the presence of the spontaneous magnetization, which is non-reciprocal in time (i.e. direction of magnetization is reversed when flow of time is reversed). The optical non-reciprocity provided by the MO phenomena is required for many unique optical applications, such as magneto-optical isolators, optical circulators or true half mirrors. Hence, we propose to design a new material providing both large MO activity and ferromagnetic state at room temperature. The material structure is based on Heusler compounds, allowing large tunability of its electronic structure and hence of its optical properties. Within this contribution, we present magneto-optical properties of several Heusler compounds, with focus to Ir2- and Co2-based Heusler compounds.
Jonáš Tokarský Molecular Modeling Based on Empirical Force Fields as a Tool for the Characterization of Nanostructures
Present lecture is devoted to characterization of various nanostructures using a combination of molecular mechanics calculations performed in Accelrys Materials Studio modeling environment with several analytical techniques such as X-ray powder diffraction analysis, thermal analysis, scanning electron microscopy, etc.
Kamil Postava, Lukáš Halagačka, Jan Chochol, Jaromír Pištora
Modeling of Periodic Grating Structures
Light interaction with periodic system enable enhancement of new phenomena of practical interest, for example nano-plasmonic effects, photonic bandgap, optical isolation, and polarization dependent light scattering. Giant nonreciprocity on magneto-plasmonic periodic structures is modeled and experimentally confirmed by spectroscopic ellipsometry. Gratings with harmonic profile of permittivity tensor is modeled using Rigorous Coupled Wave Algorithm and effective medium approximation and applied to Brillouin Light Scattering experiments.
Life Sciences
Václav Snášel Parallel Data Processing
Processing large datasets generated by research or industry represents one of the most important challenges in data mining. Ever-growing datasets present the opportunity of discovering previously unobserved patterns. Although analyzing data is the major issue of data mining, general processing of large data can be extremely time consuming. Bio-inspired computation provides a set of powerful methods and techniques based on the principles of biological, natural systems. This class of methods and techniques containing evolutionary algorithms, an ant colony, neural networks, and/or swarm intelligence complements traditional techniques of data mining and can be applied in places where the earlier approaches have encountered difficulties. Bio-inspired methods have been successfully used in various fields ranging from software engineering and data processing to chemical engineering and molecular biology. The main goal of this project is the design and utilization of powerful bio-inspired methods and algorithms for data mining that will reflect current trends in data processing and hardware evolution.
Sebastián Basterrech Time Series Modeling and Forecasting Using Reservoir Computing Models
Since the early 2000s, Reservoir Computing (RC) has gained prominence in the Neural Computation community. In a RC model there are at least two well-differentiated structures. One is a dynamical system called reservoir whichs transform the spatio-temporal information of the input data in a new larger space.This projection can enhance the linear separability of the input data.
Another structure is a memory-less learning tool designed to be robust and fast in the learning process. RC models are an alternative of Turing Machines and Recurrent Neural Networks to model cognitive processing in the neural system. Additionally, they are interesting Machine Learning tools to Time Series Modeling and Forecasting. The initialization of the reservoir parameters may influence the model performance. We analyze the impact of the model parameter in the stability of some RC models. Below, we present an empirical comparison of different reservoir initializations using a range of time series benchmarks.
Marek Penhaker Processing of EEG Data
This contribution discusses ways of measuring and processing large-volume data obtained via electroencephalography (EEG) from human subjects performing physically stressful tasks. The data was taken by the means of unique biopotential amplifiers g.USB Amp by the Gtec company capable of recording high-frequency signals of up to 18 kHz with 3µV sensitivity. To measure the biopotentials, four 16-channel g.USBAmp modules and 64 passive Ag/AgCl electrodes are used for surface EEG sensing. The measurements were performed at the University Hospital in Ostrava with regards to a precisely defined examination schema along with measuring associated physiological parameters indicating physical stress. The data was taken via a synchronous EEG video systém. During the large-volume data processing, synchronous data were correlated when mining information on the effect of physical stress on the functionality of a brain area. The result of the experiments and data processing is a number of pattern definitions that can be defined during physical stress.
21 October 2013 1:45 PM - 3:15 PM Life Sciences
21 October 2013 1:45 PM - 3:15 PM Life Sciences
Ladislav Zjavka Differential Polynomial Neural Network Applications
Lots of problems, described by real data observations, involve unknown relations of variables, which can define a differential equation based model of a complex multi-parametric function approximation. Common artificial neural network functionalities in general are based on trained similarity kinships with testing patterns. Differential polynomial neural network is a new type of neural network, which constructs and resolves an unknown general or/and ordinary partial differential equation, describing a system model with dependent variables. It is substituted with a sum of fractional relative terms defining partial derivative dependent changes of some input variables of polynomial combinations. This type of non-linear regression is not based on a whole-pattern similarity relationships and absolute values of input variables but the learned generalized data relations. This method could be a new way of modeling complex systems, which might improve standard time-series predictions as the characteristics of differential equation solutions can facilitate a greater variety of model forms than allowed by standard soft computing techniques. Brain seems to create a system of differential equations and process relative values of variables too, to define an abstract model of input pattern data relations. However neural cells supposedly apply time-delayed dynamic periodic activation functions to form relative polynomial derivative terms of differential equation solutions.
Stanislav Mišák Active Energy Management of Smart Grid
The increased penetration of renewable energy sources (RES) into the energy mix of many developed nations is important for reduction of our dependence on fossil fuels and for mitigating the contribution of power generation to climate change. However, it also brings new challenges associated with RES intermittency and the limited predictability of their availability. In addition, energy from many types of RES has high variability that can negatively affect the quality of power in the grid they are connected to. To ensure that power quality remains within acceptable limits, operational thresholds of important power quality parameters are specified by many national and international regulatory bodies. Examples of such parameters include system voltage, frequency, power factor, harmonic distortion, voltage unbalance and long and short-term flicker. Although devices that allow control of these parameters exist, the magnitude and time of their response may not be sufficient for many common operational conditions. In this contribution, we propose to enhance the capability of such devices through prediction of quality parameters of power produced by photovoltaic power plants.
Advanced Non-standard Methods
21 October 2013 3:45 PM - 5:15 PM Advanced Non-standard Methods
Petr Sosík Multi-Agent Systems Modeling and Simulation
Multi-agent systems play prominent role among the results of the Silesian University division during the last year. There are two basic lines of the multi-agent research: membrane computing (P systems, P colonies), a bio-inspired discrete computational model with various applications; the results of this kind were published in impacted journals and at several international conferences, in a form of invited or regular talks, one paper was also awarded the Best Paper Award. An interesting paper presented a membrane model simulating surface runoff, intended to predict the flooding risk. The model has been already implemented and it is highly parallelizable. Another group of results is focused on multi-agent based UAV systems, with possible applications in crisis communication and management, traffic monitoring etc. Other topics studied in the division SLU include solutions to various kinds of optimization problems.
Radomír Perzina Self-Learning Genetic Algorithm for a Timetabling Problem with Fuzzy Constraints
A Timetabling Problem is an NP-hard combinatorial optimization problem which lacks analytical solution methods. During the last two decades several algorithms have been proposed, most of which are based on heuristics like evolutionary computation methods. In this paper, to solve this problem we present a specific genetic algorithm with fuzzy constraints. Our method incorporates a self-learning genetic algorithm using indirect representation based on event priorities and heuristic local search operators to tackle real world timetabling problems. By using fuzzy sets we measure the violation of soft constraints in the fitness function in order to take care of inherent uncertainty and vagueness involved in real life data. The proposed technique satisfies all of the hard constraints of the problem and achieves a significantly better score in satisfying the soft constraints. The algorithm is computationally effective as it is demonstrated on a small, realistic example for which an optimal solution is already known due to exhaustive calculation. The structure of the algorithm enables parallel computations which are necessary for solving large real life problems.
Hashim Habiballa EQ-Algebras Generation through Genetic Algorithms EQ-Algebras Generation through Genetic Algorithms
EQ-algebras form a succesfull truth-value structure. Searching these algebras requires enourmous load to human since the state-space is superexponential with respect to its size. We present a promising method for automated generation based on Genetic Algoritms. It enables to generate interesting algebras without manual selection and evaluation of required and optional properties. We present also software application for this task - EQCreator.
Engineering
Dalibor Lukáš, Radim Blaheta
Numerical Modeling for the Solution of Engineering Problems: an Overview and Some Special Topics
The lecture overview the main directions within the research programme and concerns some special topics as e.g. solution of multiphysics problems.
Marek Lampart Attenuation of Vibrations of an Electromechanical System
A new model of an electro mechanical system damped by impact element with soft stops was developed and analyzed. This model was inspired by real frequently occurring technological problems when electromechanical rotating machines are excited by a combined loading produced by the rotor unbalance and ground vibrations. The equations of motion are solved numerically by the explicit Runge-Kutta method and solutions for given parameters are reported. Dynamics of the model is forced not only by the rotor unbalance but also by the vibration of the base plate that plays the key role here. The computational simulations proved that application of the impact body arrived at significant decrease of vibration amplitude of the rotor frame. It was observed that the amplitude of the rotor casing could be minimal for suitable values of parameters and that movement of the impact element could be chaotic. Further research in the field will operate with systems of the middle size that will be simulated by the finite element method on the new IT4Innovations supercomputer ANSELM.
Ondřej Jakl, Jiří Starý, Radim Blaheta
Homogenization, Singular Systems, HPC on the ANSELM Supercomputer
This paper deals with efficient solution of large scale symmetric positive semide nite problems which appears in the solution of multiscale problems and numerical upscaling. The lecture will start with an overview of upscaling or homogenization techniques solved within the RP2 research programme. Then it focuses on upscaling for analysis of geocomposites including solution of problems with pure Dirichlet and pure Neumann boundary conditions. These problems are solved by sw GEM developed at the Institute of Geonics AS CR. Efficiency of computations at different parallel computers including the new IT4Innovations supercomputer ANSELM.
Radek Kučera Algorithms for Solving Nonlinear Problems of Contact Mechanics
The lecture overview some algorithms for solving contact problems with friction as the algorithm of active sets, nonsmooth Newton‘s method, interior point algorithm and relaxation type algorithm using natural decomposition the bodies in contact. The achieved convergence results will be described for the individual algorithms. Computational efficiency and applicability of algorithms will be demonstrated at the solution of model problems. A number of theoretical questions as well as some implementation aspects remain as open problems and motivation for a further research.
22 October 2013 9:00 AM - 10:30 AM Engineering
22 October 2013 11:00 AM - 12:30 PM Engineering
Tomáš Kozubek About Us
Our research team has been established on September 1st 2011 along with other teams and research programs at IT4Innovations. Our work is based on successful research carried out at Department of Applied Mathematics at VSB TU Ostrava in area of scalable numerical algorithms. Together with Department of Mechanics and Department of Mechanics of Materials these novel algorithms have been effectively used to solve challenging engineering problems. We have also the advantage of long-term collaboration with renowned domestic and international research centers and industrial partners. The members of our team are mostly recruited from the employees of departments listed above. In many cases our researches are still active in teaching, training and advising our talented students. Nevertheless our primal goal is to help our colleagues at IT4I, members of academic community of the Czech Republic and our industrial partners with parallel computing. Thanks to our skills and knowledge everybody who is interested can use modern parallel methods to solve their practical problems without deep knowledge of high performance computing and parallel programming. We are open to collaboration with industry or academia. Our expertise is proven by number of successful projects. For more details see our web: http://industry.it4i.cz
Tomáš Karásek Supercomputing for Industry
One of the primary goals of Research Programme Libraries for Parallel Computing is collaboration with industrial partners. In this presentation several topics will be presented. First capabilities of IT4I in area of Computational Fluid Dynamics (CFD) and Computational Structural Dynamics (CSD) for solving real industrial problems. Examples of solved problems aplicable in automotive, aerospace, energy, engineering and environmental modeling will be presented. Next software available at IT4I for solving real industrial problems will be presented. In this part overview of both commercial and open source solutions for high performance computing will be given. Last but not least possibility of collaboration modes such as commercial and joint research will be discussed.
David Horák Comparison of FLLOP‘s Parallel Solvers of Linear Systems Based on Domain Decomposition
Large sparse linear systems of equations can be solved by means of direct, iterative or hybrid solvers applied to the original or transformed primal, mixed or dual problem. The presentation deals with the applications of non-overlapping domain decomposition (DD) techniques implemented in our novel FLLOP library for their solution and comparison. The resulting problem can be solved iteratively by the conjugate gradient (CG) method in primal variables or after transformation by the combination of CG with direct solvers in dual variables (FETI). Grouping of the nodes based on DD can be even used to generate the subspace for deflated CG (DCG) applied to the undecomposed problem. Numerical experiments illustrate the efficiency of parallel implementations of various methods on several architectures.
René Kalus Molecular Dynamics Group - 2013 Survey and 2014 Plans
Survey of the MOLDYN group work carried out within the period of November 2012 through October 2013 will be presented for the main branches of the group research focus, namely, a) simulations of non-adiabatic dynamics of open-shell atomic clusters and b) classical Monte Carlo simulations of small and medium-sized molecular clusters. Prospects of novel branches of research which are currently being developed in the group, quantum Monte Carlo simulations and clusters structures optimizations, as well as prospects for the next year will also be reported.
Audio and Video Processing
22 October 2013 1:30 PM - 3:00 PM Audio and Video Processing
Markéta Dubská Object Detection and Vanishing Points
This paper deals with the detection of orthogonal vanishing points. The aim is to efficiently cope with the clutter edges in real-life images and to determine the camera orientation in the Manhattan world reliably. We are using a modified scheme of the Cascaded Hough Transform where only one Hough space is accumulated. The parameterization of the vanishing points is based on the line parameterization using parallel coordinates and it is defined as a mapping of the whole real projective plane to a finite space. The parameterization is in all aspects linear; it involves no goniometric or other non-linear operations and thus it is suitable for implementation in embedded chips and circuitry.
Martin Hlosta Constrained Classification of Large Imbalanced Data
The talk deals with classification of highly imbalanced data with accuracy constraints for the minority class. We solve this problem by meta-learning method that uses cost-sensitive logistic regression (LR) to provide initial solutions for two optimization algorithms - Genetic Algorithm and Particle Swarm Optimization. This leads to better results than separate usage of LR and optimization algorithms.
Petr Zemek Jumping Finite Automata
We propose a new investigation area in automata theory: jumping finite automata. These automata work like classical finite automata except that they read input words discontinuously; that is, after reading a symbol, they can jump over some symbols within the words and continue their computation from there. We establish several results concerning jumping finite automata in terms of commonly investigated areas of automata theory. Most importantly, we give several results that demonstrate differences between jumping finite automata and classical finite automata.
Embedded Systems
Lukáš Sekanina Evolutionary Design and Optimization
This presentation surveys our recent results in evolutionary design and optimization of digital circuits. Case studies include evolutionary design of switching image filters and approximate arithmetic circuits. A new platform for evolution and adaptation of image filters which is based on dynamic partial reconfiguration of FPGAs will be presented. Last part of the talk will be devoted to computational intelligence in traffic simulation and prediction.
Jan Kořenek Hardware Acceleration in Computer Networks
With the growing speed of computer networks, network devices need more processing power to achieve wire speed throughput. As current processors have limited performance, routers and other network devices use hardware acceleration to achieve wire speed throughput with reasonable power consumption. Usually, the throughput is decreased by time critical operations which have to be performed for every packet or every byte of network traffic. The presentation provides our research results in hardware acceleration of time critical operations and introduces current research projects and cooperation with commercial companies.
Martin Drahanský Biometric Security Systems: Liveness Detection, Skin Diseases on Fingers, 3D Hand and Face, Eye Iris and Retina
The presentation will be oriented on biometric security systems with the following specialized topics: liveness detection - recognition whether the presented biometric feature comes from living and genuine user; elimination of the influence of skin diseases on fingertip to the fingerprint recognition process; recognition of persons based on their 3D hand and face geometry and features extracted from eye iris and retina.
22 October 2013 3:30 PM - 5:00 PM Embedded Systems
www.it4i.cz
