Spring 2017 Programme

Spring 2017 Programme

This is the programme of the Sitola seminar in spring 2017. Presentations for the current semester are available here.

  • 22.2.2017
    Mgr. Michal Zima
    Towards sustainable blockchain
    Abstract: Bitcoin, a digital currency—cryptocurrency, not only started a new era of digital money, but also came with a technology for trustless distributed databases: a blockchain. As originally designed and as used nowadays, blockchains are ever-growing databases, storing all data ever saved into them. However, this property questions sustainability of such blockchains as well as their general usability.
    In this talk, we present and discuss today’s approaches to this issue and offer our alternative methods which lead towards a blockchain of a bounded size.

  • 1.3.2017
    Dr. Tanvir Shaikh
    Image analysis of electron micrographs using SPIDER
    Abstract: Using the electron microscope, one can reconstruct a model of a three-dimensional object from two-dimensional images to near-atomic resolution. Biological specimens are especially challenging because they are beam-sensitive, and so hundreds, thousands, or sometimes even millions of images must be combined to get enough signal.
    SPIDER is one of the oldest software packages used for this purpose, in use since 1978. I will describe old and new challenges in the field, and what strategies we might use to solve these challenges.

  • 8.3.2017
    Ondřej Bilčík
    Software-defined networking for university infrastructure
    Abstract: Software-defined networking (SDN) is a modern approach to computer networking that allows network administrators to build flexible and agile networks that quickly adapt to dynamically changing needs. It works by separating the network’s control plane and the data plane (physical devices), creating software-programmable, centralized and extensible orchestration platform that can automate the provisioning and configuration of the entire infrastructure. Besides basic SDN principles, our talk will present real-life examples of applying SDN concept — specifically SDN/OpenFlow — to complex infrastructure like university network, illustrating the key SDN benefits by their comparison with traditional approach.

  • 15.3.2017
    Mgr. Adam Midlik
    Protein secondary structure annotation – recognition of protein’s skeleton
    Abstract: Protein structural data are currently one of the most interesting and important research resources. Nowadays, we have available even rich datasets about individual protein families (groups of proteins with similar structure). To examine them, we have to orient within the collected data. Specifically, to identify and describe key regions in different proteins from one protein family (binding sites, channels, etc.). For these purposes, it is essential to identify firm parts of the proteins – their secondary structure elements (helices, sheets), which form something like a protein’s skeleton.
    In this presentation, I will show an algorithm for identification and annotation of protein secondary structure elements and its implementation. Afterwards, I will also demonstrate its functionality on an example of important protein family (cytochromes P450, which are responsible for degradation of drugs and other molecules coming to organism from outside).

  • 22.3.2017
    RNDr. BcA. Tereza Pařilová, DiS., MBA
    Individualism in assistive technologies – innovative way of its use
    Abstract: Assistive technologies were always mostly used in domain of deaf or blind users. A huge gap is identified in assistive technologies aiming at users with neurological disorders like ADHD or dyslexia. Nowadays there is growing number of such users that need adequate technological solution enabling them to access and understand written information correctly. Reading is critical for correct cognitive development, social integration and last but not least for educational opportunities. Such users ask for strong individualism, so maximal personalization of application is needed.
    The presentation will mention tools used by dyslexics in general and introduce problems that dyslexic users are facing with missing technologies. However, the presentation will be focused on our Dystexia project that designs, invents and tests fully automated customizing web service. The project uses existing technological and mathematical approaches in innovative ways that make the service the first one really aiming at dyslexic users’ specific needs.

  • 29.3.2017
    prof. RNDr. Michal Kozubek, Ph.D.
    Toward collaborative environment for developing, benchmarking and searching bioimage analysis methods
    Abstract: The bioimaging community has recently realized the strong need for benchmarking various image analysis methods and software packages in order to compare their performance and assess their suitability for specific applications. This presentation summarizes the recent progress in this respect, reviews available benchmarks and describes the general process of designing a bioimage analysis benchmark or challenge (benchmarking effort associated to a particular conference), including proper selection of reference image datasets (both simulated with precisely known ground truth as well as real ones annotated by experts) and evaluation metrics. Examples of biological applications will be shown for which benchmarking of relevant image analysis methods have already been performed. Finally, attention will be paid to the development of web-based and cloud-based resources for developing, benchmarking and searching bioimage analysis methods.

  • 5.4.2017
    Discussion about future SITOLA

  • 12.4.2017
    Mgr. Lukáš Ručka
    Fast reconstruction of image deformation field
    Abstract: Computer-aided medical analysis is the key component of current invasive treatment. With aid of computer simulations deployed as part of intraoperative navigation, it is possible to expose the patient to less harmful examinations during surgeries. The key component of intraoperative navigation is the update of precise, pre-operative scans of patient’s body according to current conditions and situation, as perceived during the surgery. To facilitate such updates in near real-time, a simplified (sparse) representation of the image data has to be considered. However, the discrete and sparse image deformations developed upon such data cannot be applied to the original image domain. To overcome this issue, a suitable inter(extra)polation technique has to be deployed.
    This presentation covers the basics of deformation extrapolation from computer model in the context of medical imaging. Afterwards, the audience gets acquainted with radial-basis functions interpolation framework and its properties. Following, results of radial-basis functions interpolation are evaluated over internal organs of quadruped test subject.

  • 19.4.2017 15:00
    Flash talks with 5 minutes for presentation and about 2 minutes for question(s) per person

    Management interface variation in heterogeneous cloud federations
    Mgr. Boris Parák
    Abstract: This talk takes a brief look at the variation in IaaS management interfaces in Private-Private, Private-Public, and Public-Public heterogeneous cloud federations. Identified differences are broadly categorized based on federation type, character of the difference, and estimated mitigation complexity.

    Very high quality and low latency audiovisual transmissions
    RNDr. Miloš Liška, Ph.D.
    Abstract: We provide an overview of the latest development and research results in the field of very high quality and low latency audiovisual transmissions and its deployments for most interesting use cases in different user communities.

    On scheduling mixed workloads in shared cloud infrastructures
    RNDr. Dalibor Klusáček, Ph.D.
    Abstract: Modern computing services such as clouds, grids or HPC clusters are both complex and costly installations. To utilize them properly, competing frameworks/workloads must share the same underlying infrastructure in order to avoid resource fragmentation and poor utilization, requiring complex scheduling mechanism. In this talk we use real life-based observations from the CERIT Scientific Cloud to show some of the scheduling challenges that we are facing today.

    Hotspot searching in scientific software
    Mgr. et Mgr. Jaroslav Olha
    Abstract: Scientific software tends to be computationally intensive and its long runtimes can be frequently exacerbated by poorly-optimized code, which creates needless performance bottlenecks. A case study will be used to briefly demonstrate the power of hotspot searching, and to show how simple performance inefficiencies can be found and fixed to dramatically improve the performance of such programs.

    Integration of buses into the traffic signal control
    Vinh Quang Dang, Ph.D.
    Abstract: Traffic signal control in a network of signalized intersections is an important practical problem. We briefly present an intersection control problem considering bus integration formulated as a machine scheduling problem, and a solution idea to address this problem based on a schedule-driven intersection control algorithm using forward search.

    Autotuning OpenCL kernels
    RNDr. Jiří Filipovič, Ph.D.
    Abstract: First, we demonstrate possible improvements of performance and portability of the code (OpenCL in this particular case) by autotuning. Second, we outline research opportunities in autotuning and challenges we are targeting with our autotuning tool.

  • 26.4.2017
    doc. RNDr. Vlastislav Dohnal, Ph.D.
    Optimizing Performance of Query Evaluation in Metric Spaces
    Abstract: Similarity searching has become widely available in many on-line archives of multimedia data. Users accessing such systems are looking for data items similar to their specific query needs, which also follow time-related trends, e.g. currently happening events or natural wonders. Existing index structures organize data-sets in a homogeneous and query-agnostic way, so they do not adapt to the users‘ constantly changing preferences.
    In this talk, we report on techniques to improve organization of priority queue of data partitions, which is inherently used in a k-nearest neighbor query evaluation algorithm. We will also present result of experiments.

  • 3.5.2017
    RNDr. Tomáš Rebok, Ph.D.
    (Big) data analysis for purposes of police investigations
    Abstract: During our talk, we will present current results of our collaboration with the Police of the Czech Republic, focused on (big) heterogeneous data analysis related to real crime investigations. We will present our experience and identify main issues of such analysis as well as introduce tools and technologies employed for the prototype implementation of developed CopAS (Cops Analytic System) tool. Objectives and goals of related research project, supported by the Ministry of Interior of the CR, will be also presented during the talk.

  • 10.5.2017
    Bc. Matej Minárik, RNDr. Miloš Liška, Ph.D.
    Automated anonymization of modalities for medical video transmissions
    Abstract: High quality and low latency video transmissions are successfully being deployed for live cases transmission for educational purposes to medical workshops, conferences and congresses. A key requirement is to transmit and offer the audience the complete overview of the case in the same way as it is available to the medical personnel within the operating theater. In cases such as the cardiologic or gastroenterologic operations this also means to transmit possibly all the modalities (various types of equipment or probes used to acquire images of the body) needed by the medical personnel. Images or videos produced by such an equipment contain patients personal data hard coded within the images for the purpose of its simple and clear identification. Such a practice is however undesirable during the live case transmission for educational purposes where the patient has to stay completely anonymous and unidentified by the audience.
    In this talk we present out approach to automated anonymization of said modalities for the purpose of high quality and low latency video transmissions. We cover topics ranging from optical character recognition and image processing including GPU acceleration to semantic analysis of the textual patients personal data. We also provide an insight into the performance characteristics of the automated anonymization process we have designed and implemented.

  • 17.5.2017
    Ing. Eva Výtvarová
    Community detection algorithms in temporal brain networks
    Abstract: The functional connectivity of human brain can be seen as a reflection of emergent effects of dynamic processes on structural connectivity, specifically the spatio-temporal patterns of brain activity. These patterns / communities can be detected by many methods, one of the most widely used is modularity. I am going to show the effect of random matrix theory decomposition as a step preceding community detection. Further, the results from applying different modularity algorithms to synthetic network and to connectivity network derived from time-series of observations of dynamic system will be discussed. I will also talk about computational complexity of particular pipeline steps.

  • Examination period 23.5. – 30.6.2017
    Meetings devoted to state exam training below are scheduled mostly. All state exam trainings are in Czech. More days can be added if needed by students.

  • 7.6.2017
    State exam training

    David Štípský: Odhad doby zastavení autobusu
    Vedoucí: Hana Rudová
    Oponent: Lubomír Popelínský

    Roman Lacko: Rozvrhování státních zkoušek na Fakultě informatiky
    Vedoucí: Hana Rudová
    Oponent: Pavel Matula

  • 8.6.2017 Thursday 15:00
    State exam training

    Ondřej Hlouša: Rozvrhování zaměstnanců ve vědeckém zábavním centru
    Vedoucí: Hana Rudová
    Oponent: Dalibor Klusáček

    Matej Minárik: Akcelerace automatické anonymizace modalit
    Vedoucí: Miloš Liška
    Oponent: Martin Pulec

  • 14.6.2017
    State exam training

    Pavel Zvonař: Optimalizace a řízení spotřeby el. proudu v domácnosti
    Vedoucí: Eva Hladká
    Oponent: Dalibor Klusáček

    Petra Ondřejková: Pokročilé metody modelování elastických deformací v prostředí Matlab
    Vedoucí: Igor Peterlík
    Oponent: Jan Fousek

  • 21.6.2017
    State exam training

    Lukáš Kotol: Systém pro zpracování provozních statistik videokonferenční infrastruktury
    Vedoucí: Miloš Liška
    Oponent: Jan Růžička

    Marek Perichta: Správa zdrojov pre rezervačný systém Shongo
    Vedoucí: Miloš Liška
    Oponent: Ondřej Pavelka

    RNDr. Jiří Filipovič, Ph.D.
    Autotuning techniques for accelerators
    (16:00, survey, 45 minutes presentation)

    Abstract: Performance optimization is important step in application development, especially in HPC area. With growing complexity, heterogeneity and parallelism of nowadays hardware, achieving high performance is a difficult task. Moreover, optimization steps are usually hardware-dependent and thus new generations of hardware may require to re-optimize the software. Therefore, automatic search for efficient program parameters is of high importance, as it eases software development process and allows developers to write software, which is able to re-optimize itself for a new architecture.
    In this talk, I will introduce various autotuning techniques, targeting many performance-critical aspects (e.g. compiler flags selection, re-configutation of parallel communication, or generation of various implementation and empirical search for the fastest one). I will also sketch which areas are of primary interest of HPC group in CERIT-SC.

Contact: Hana Rudová

(c) 2018 SITOLA, Administration