Autumn 2017 Programme

Autumn 2017 Programme

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

  • 20.9.2017
    RNDr. Jan Fousek
    Efficient sparse matrix-delayed vector multiplication for discretized neural field model
    Abstract: Computational models of the human brain provide an important tool for studying the principles behind brain function and disease. To achieve whole-brain simulation, models are formulated at the level of neuronal populations as systems of delayed differential equations. In this presentation, we show, that the integration of large systems of sparsely connected neural masses is similar to well studied sparse matrix-vector multiplication, however, due to delayed contributions it differs in the data access pattern to the vectors. To improve data locality we propose a combination of node reordering and tiled schedules derived from the connectivity matrix of the particular system, which allow performing multiple integration steps within a tile. We present two schedules: with a serial processing of the tiles and one allowing for parallel processing of the tiles. We evaluate the presented schedules showing speedup up to 2x on single-socket CPU, and 1.25x on Xeon Phi accelerator.

  • 27.9.2017
    RNDr. Vít Rusňák, Ph.D.
    Multi-scale data navigation in tabletop environments
    Abstract: Multi-scale data such as geo-imagery or large graph exploration require users to interact with elements at various levels of precision and details. Previous work introduced multiple techniques for data exploration such as various types of interactive lenses. However, they are designed for single-user scenarios and their adaptation in multi-user co-located collaborative scenarios is troublesome. In the talk, we present our ongoing work in this area with focus on co-located collaboration. We illustrate the problem of co-located collaborative work of multiple users on a prototype of geo-imagery application for multi-touch tabletops. Next, we propose the set of interaction actions and we introduce two interface designs for multi-scale data explorations and their evaluation.

  • 4.10.2017
    Ing. Matej Lexa, Ph.D.
    Algorithms and machine learning in the study of non-canonical DNA structures
    Abstract: While DNA is typically represented in textbooks as a helix of two antiparallel strands, in real world this molecule often forms alternative structures (e.g. triplex and quadruplex DNA). Although the relationship between DNA sequence and the capacity to form such structures has been known for some time qualitatively, the exact rules for their formation are not known. Together with colleagues from Brno Technical University we improved the procedure for searching for sequences that are likely to form non-canonical DNA structures. The new methods were published as R/Bioconductor packages. User can now easily search genomes for potential triplex- and quadruplex-forming sequences and use them in downstream analyses. I will provide background information, show our solutions and demonstrate the use of the packages.

  • 11.10.2017
    Ing. Vlad Popovici, M.Sc., Ph.D.
    Multimodal bioinformatics: even bigger data
    Abstract: Hight throughput genomic revolution started almost twenty years ago with the first in-house printed DNA chips. Since then, various technologies evolved, allowing the interogation of the whole (human) genome, proteome, metabolome, etc., all producing large amounts of data. Bioinformatics tools and methods evolved to account for all these data modalities with the current bottleneck being the integration of these perspectives into a more comprehensive picture. In parallel and completely independent of bioinformatics, digital pathology also witnessed significant advances fuelled mostly by technological developments: slide scanners and computational infrastructure. However, both „classical“ bioinformatics and digital pathology/bioimaging are often used to investigate the same biological phenomenon. It is, therefore, natural to attempt to combine these two seemingly incompatible fields with the hope of unveiling new connections between them. In this talk we will look at three examples of jointly mining the transcriptome and the histopathology images in the context of breast and colon cancers. We will also discuss the computational challenges one faces when working with these data.

  • 18.10.2017
    Mgr. Ivana Hutařová Vařeková
    Mapping and 2D visualization of secondary structure elements in proteins
    Abstract: Data about molecular structures growths rapidly each day and provides a rich and highly interesting resource for IT driven life science research. An important branch of this data is information about proteins and their families – groups of proteins having the same biological function. For discovery of protein families, we require 2D scheme, which transparently depicts structures of proteins within an individual family and allows their visual investigation. A suitable way is to use a SSE 2D diagram – a scheme showing protein secondary structure elements (SSEs) in 2D. This diagram must have as low bias of SSEs distances (in comparison with 3D structure) as possible. We present an approach, which is able to produce such SSE 2D diagrams and we show its applicability on cytochrome P450 protein family.

  • 25.10.2017
    doc. Ing. RNDr. Barbora Bühnová, Ph.D.
    Software architecture optimization in the context of PaaS Cloud
    Abstract: Software architecture design is one of the key activities in any software engineering process. The decisions made during software architecture design have significant implications on quality goals related to the developed software product. To better guide the software architect along the design process, various architectural tactics and other support is being investigated. In the talk, I will give an overview of the area and discuss our progress in guiding software architects during the design of PaaS (Platform as a Service) Cloud applications. This can hardly be done without effective tool support generating application prototypes and evaluating them in the target environment, which is our current work.

  • 1.11.2017
    Ing. Jana Pazúriková
    Accelerating Chemical Simulation through Model Modification (thesis defense rehearsal)
    Abstract: In the first part of my talk, I will propose a methodology to deal with demanding computations. A lot of them share a common concept: the evaluation of a computationally expensive function in a loop. The usual approaches aim to accelerate the function, but they are beginning to hit the limits. We, on the contrary, turn our attention to the loop. First, we build a model of the problem that emphasizes the iterative character of computation. Then we modify the execution of the loop in various ways that lead to the acceleration of the whole computation. We apply this methodology to three demanding computational problems from chemical simulations: two regarding their long timescales and one regarding the accuracy. We change the scheme of the loop to add another level of parallelization, we reduce the number of the function’s evaluations by reusing calculations from previous iterations or we omit iterations that do not contribute to the result.
    In the second part of my talk, I will explain the methodology in better detail on one example. We approximated the calculation of the mean square distance between many molecular structures, a demanding part of metadynamics simulations. Metadynamics pushes the molecular dynamics simulation forward to quicken the occurrence of rare events. Our method manages to reduce its overhead significantly without a loss of accuracy.

  • 8.11.2017
    RNDr. Jiří Filipovič, Ph.D.
    Acceleration of 3D reconstruction in cryo-EM
    Abstract: Cryo-electron microscopy (cryo-EM) gains very high popularity in recent years. It studies specimens in their natural environment at near-atomic resolution. However, it requires high computational power to get a 3D structure from raw data generated by a microscope (processing one specimen requires at least thousands of CPU hours).
    A 3D reconstruction is one of main bottlenecks in reconstruction of cryo-EM data. During the 3D reconstruction, multiple 2D images are placed into 3D space with estimated rotation, creating a 3D voxel structure. There are several existing GPU implementations of 3D reconstruction, implementing scatter pattern. The scatter pattern, however, has poor memory locality and must handle race conditions. In comparison to state-of-the-art, our algorithm uses gather pattern, which exposes better memory locality and minimizes number of atomic accesses into memory. Moreover, we have introduced approximated algorithm, which introduces negligible error in output and is an order of magnitude faster. Our algorithms are implemented in Xmipp, a widely-used software for cryo-EM.
    In this talk, I will introduce our algorithms, their GPU implementations and discuss some programming challenges related to the topic.

  • 15.11.2017
    doc. RNDr. Petr Holub, Ph.D.
    Implemented FAIR and FAIR-Health principles in BBMRI-ERIC, a large European distributed biomedical infrastructure
    Abstract: Medical research is moving fast forward delivering new treatments for patients and strategies for public health. However, it is also facing a number of substantial challenges around reproducibility (estimates ranging between 50-90% of medical research being irreproducible!) and findability, accessibility, interoperability and reusability (FAIR principles) of resources such as research data and biological material. Another important aspect is data security and privacy protection, as medical research typically deals with personal data. In BBMRI-ERIC, Biobanking and Biomolecular European Research Infrastructure, one of the largest medical research infrastructures in Europe, we are striving to address these challenges for biological material and associated personal data and biomolecular resources. The FAIR principles will be introduced as well as their FAIR-Health extensions, together with showcase of their implementation within BBMRI-ERIC.

  • 22.11.2017
    Mgr. Eva Budinská, Ph.D.
    Big data of molecular biology experiments in cohort studies
    Abstract: State-of-the-art high-resolution molecular technologies as applied in current biomedical research generate huge amounts of data. Each technology produces specific data types and their pre-processing and analysis requires tailored analytical and algorithmic solutions. In this talk I will introduce a new cohort study CELSPAC-TNG and will discuss the bioinformatics challenges our team is facing. In particular, I will focus on processing of data from high-resolution mass spectrometry experiments used for analysis of chemicals and metabolites.

  • 29.11.2017

  • 6.12.2017
    Ing. Jiří Jaroš, Ph.D.
    Acceleration of Ultrasound Simulations
    Abstract: Ultrasound simulations have very broad area of practical applications in many fields ranging from geological exploration, non-destructive material testing, design of ultrasound transducers and sensors, to personalized medicine. All these fields require precise physical and numerical models to be applied on extensive simulation domains. Unfortunately, these modes suffer from very high computational requirements. Our group has been involved in the development of efficient decomposition and parallelization of the ultrasound models. By introducing the Local Fourier Basis decomposition of the k-space pseudospectral methods in the k-Wave toolbox, the computational burden of these modes has been significantly reduced and over 8000 registered users have been attracted to use this toolbox. This talk will introduce the current HPC model for ultrasound simulations and discuss open challenges on the way to routine routine industrial and medical use.

  • 13.12.2017
    RNDr. Dalibor Klusáček, Ph.D.
    Scheduling Challenges in a Shared Private Cloud Infrastructure
    Abstract: Modern computing services such as clouds, grids or HPC clusters are both complex and costly installations. Therefore, it is a major challenge to utilize them properly. At the same time, the variety of provided services implies that it is no longer suitable to host each service separately using a fully dedicated hardware. Instead, competing frameworks/workloads must share the same underlying infrastructure in order to avoid resource fragmentation and poor utilization. This simultaneous execution of different workloads with different needs and requirements represents rather nontrivial scheduling challenges. In this talk we analyze a real-life example of such an installation using the data and experience from the MetaCentrum infrastructure. We also present some of the scheduling techniques used to solve the challenging problems observed in MetaCentrum.

    and Christmas meeting

  • Examination period 2.1. – 16.2.2018
    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.

  • 24.1.2018
    State exam training
    Ludík Hájek: Rozvrhování pro zážitkovou cestovní kancelář
    Vedoucí: Hana Rudová
    Oponent: Filip Nguyen

  • 31.1.2018
    State exam training
    Jozef Pastucha: Webová aplikace pro vizualizaci trendů kvality biomakromolekul a ligandů
    Vedoucí: Tomáš Raček
    Oponent: Radka Svobodová Vařeková

    Dan Polanský: Webová aplikace pro analýzu dat z databáze
    Vedoucí: Tomáš Raček
    Oponent: Radka Svobodová Vařeková

  • 7.2.2018
    State exam training
    Bartoňková Lucie: Netradiční výukové hry pro výuku matematiky
    Vedoucí: Radek Pelánek
    Oponent: Jiří Řihák

    Filip Obšivač: Implementace vybraných empirických metod pro výpočet parciálních atomových nábojů
    Vedoucí: Tomáš Raček
    Oponent: Vladimír Horský
  • Contact: Hana Rudová

    (c) 2018 SITOLA, Administration