Research on high performance computing was initiated by
the first Director of the Institute of Computer Science
Professor Jacek Mościński in early 80-ties.
Nowadays, our activity is concentrated on:
- Data Science
- Efficient Distributed Data Access and Cloud Technology
- Ontological Representation of Grid Resources
- Problem Solving Environments
- Computer Models of Reality and High-performance Simulation
- Pattern Recognition and Intelligent Data Exploration
- Modeling Natural Phenomena for Computer Graphics and Animation
- Computational Social Choice and Complexity of Elections
Since the works of Hollerith in 1890s, machines have been used to aid humans in processing of vast amounts of data. Nowadays, as we explore natural, technological and business processes for which analytical models are not yet uncovered, the focus is shifted towards intelligent data recognition, analysis and understanding techniques.
At the Department of Computer Science, we have gained proficiency in developing and applying pattern recognition solutions. Our expertise includes methods for automatic classification and clustering, by using supervised and unsupervised learning coupled with advanced noise removal techniques. Our tools for data visualisation and dimensionality reduction provide intuitive support for data analysts. We also excel in automatic information extraction with image processing algorithms and interaction network analysis.
Main application schemes of our methods include intelligent system control, exploratory data analysis and decision support. Our clustering techniques have been used for surveillance and safe-guarding of IBR2 pulsed nuclear reactor and in supervising high-performance computer simulations. Ensemble classification methods of ours have been employed in an industrial drug discovery support system. Our algorithms have also proven successful in handling pattern recognition problems emerging from life sciences, e.g. genomic and proteomic data exploration, as well as cancer detection.
Our activities have been funded by grants from Polish State Agencies: KBN in 1993-95, MNiI in 2004-06 and MEiN in 2005-07, as well as funds from the University of Minnesota Digital Technology Center in 2003-04.
We collaborate on extending, customising and applying our algorithms with leading foreign and national research and industry partners, including:
- Minnesota Supercomputing Institute, – computer simulation results analysis,
- University of Minnesota Medical School, USA – drug discovery, cancer patterns analysis,
- University of Southern California, USA – earthquake patterns exploration,
- Joint Institute for Nuclear Research, Russia – nuclear reactor surveillance,
- Fujitsu Group, Poland – drug discovery support,
- Collegium Medicum of the Jagiellonian University, Poland – cancer detection.
Research team: Witold Dzwinel, Krzysztof Boryczko, Tomasz Arodź, Marcin Kurdziel
Modern Grid solutions provide now variety of middleware solutions that are supposed to help manage resources available in the Grid environment, but most of these solutions are based on specific formats for each aspect of the Grid, thus making inconvenient integration and making complete virtualization of resources difficult. Our response to needs for a more uniform way of defining metadata in the Grid environment is using ontologies as a formalism for unified representation of various kinds of metadata in the Grid. In the context of automatic workflow composition from semantically annotated Grid services, an extensible and distributed framework for management of semantic metadata in Grid environment called Grid Organizational Memory (GOM) was developed.
The objective was to define knowledge representation paying attention to the distributed nature of a working environment and allowing collaborative development of ontologies within any application domain. The result was the separation schema for the GOM ontologies. GOM contains the prototype versions of generic ontologies, domain specific ontologies and supports management of ontological registries. The framework architecture also reflects separation of Grid on Virtual Organizations (VO). GOM prototype is available in terms of public license. It is being used for development of ontology similarity for VO development and semi-automatic construction of workflow-type grid applications.
The scientific research in the framework of EU IST-2002-511385 K-WfGrid project and KBN Grant is being performed in tight cooperation with:
- Fraunhofer Institute for Computer Architecture and Software Technology, Berlin
- Institute for Computer Science, University of Innsbruck
- Institute of Informatics of the Slovak Academy of Sciences, Bratislava
Researchers: R. Słota, M. Majewska, M. Kuta, S. Polak, J, Kitowski with ACK Cyfronet AGH collaborators: B. Kryza, Ł. Dutka.
Usage of Grid Organizational Memory for workflow composition
Modeling of natural phenomena and processes such as behavior of water surface with splashes, forming of ocean waves, smoke and clouds is an intriguing and challenging task. In order to visualize them in realistic way in real time we need to employ simplified physical models, as well as non-physical approach. Special algorithms utilizing capabilities of modern Graphics Processing Units has to be developed to achieve desirable speed of both simulation and animation. We have implemented algorithms for visualization of fluid flow, based on continuous models in which simplified Navier-Stokes equations are solved. We also use particle methods as well as cellular automata for evolution and visualization of clouds. Non-physical models have been adopted for ocean waves simulation.
Turbulent flow through the dam visualized in real time
Real time animation of ocean waves
Volumetric effects in clouds evolution
Researchers: W.Alda, K.Boryczko, W.Dzwinel, R. Górecki, J.Kitowski
In the recent years we have been tackling problems of automated two- and three-dimensional mesh generation and adaptation. Mesh generation can be described as approximate covering of a domain with a finite number of simple adjacent elements, forming the mesh. Such meshes are commonly used in simulation of physical processes (e.g. FEM), but also in other areas, such as computer graphics, geodesy or computational geometry.
The developed generator is capable of creating unstructured triangular and quadrilateral meshes on planes or 3D surfaces, and tetra- or hexahedral volumetric meshes. For hexahedral meshes we investigate two types of methods: indirect (based on tetrahedral mesh, which has to be created earlier) and direct, using Medial Axis Transform.
We investigate also the problem of mesh adaptation for the considered model – i.e. their optimization while keeping the number of elements low, with respect to both domain geometry and simulated process. In the latter case the goal of adaptation is to obtain the best possible precision of results while reducing the computation cost (anisotropic meshes are also used). The developed methods allow for multi-criteria adaptation of meshes in order to conform to requirements obtained from different sources.
Recently, we have been investigating the problem of parallelization of the mesh generation process. It is mainly caused by the still growing sizes of used meshes. In such cases the sequential generation poses problems regarding both the computational and memory cost. This subject is developed in cooperation with Universite de Technologie de Compiegnein France.
In all problems mentioned above the particular stress is put on the efficiency and reliability of algorithms and proper selection of data structures. The important factors are also portability to different platforms and extendability of algorithms and data structures to constantly growing possibilities of the generator.
Researchers: B.Głut, T.Jurczyk, K. Boryczko, J.Kitowski