Spring 2022 Program

This is the program of the Sitola seminar in spring 2022. Presentations for the current semester are available here.

• 16.2.2022
  Mgr. Dominik František Bučík
  Authentication and authorization in large distributed infrastructures
  Abstract: While the majority of life science services are openly accessible to anyone across the world, many of them require researchers to authenticate. Sensitive data and licensed resources often require strong security for access. Research services have usually implemented local access management solutions and issued their own usernames and passwords. As a consequence, researchers quickly became overloaded with having to remember numerous login credentials. This situation has led to the development of a concept of AAI, which helps to solve this problem.
  The presentation will first provide an introduction to the AAI topic, and then it will focus on AAI in international research infrastructures, namely ELIXIR. The second half of the presentation will provide a more technical overview of selected AAI features.
• 23.2.2022
  RNDr. Igor Peterlík, Ph.D.
  When radiation saves lives: Concepts and algorithms behind modern radiotherapy
  Abstract: Radiation therapy has become an essential weapon in our fight against cancer. According to the American Society of Radiation Oncology, two out of three cancer patients receive radiotherapy— either alone or combined with other treatment approaches such as surgery or chemotherapy.
  During a radiotherapy treatment, a patient is positioned inside a machine equipped with a particle accelerator generating a high-energy beam targeting the tumor. While the dose of ionizing radiation deposited inside the cancerous tissue should be maximized, it is necessary to protect the surrounding healthy tissues and nearby organs at risk. To achieve this complex goal, modern treatment machines capable of creating dynamic custom beam shapes and beaming trajectories are controlled by advanced algorithms combining physics-based modeling, optimization, and image processing.
  After introducing the basic concepts of radiation therapy, I will focus on algorithms behind generating a patient-specific treatment plan and its accurate delivery. I will describe image-guided radiation therapy as well as the recently-introduced adaptive therapy relying on deep learning. I will conclude my talk with perspectives and emerging trends in this exciting domain.
  Short Bio: After obtaining a Ph.D. in computer science at Masaryk University, Igor Peterlik spent several years as a research fellow in France and Canada. In 2015, he obtained a tenured position at the French national institute Inria. Currently, he is a senior research scientist at Varian Medical Systems, a world leader in radiation therapy, and lives with his family in Switzerland. His research interests include physics-based simulation and modeling, numerical optimization, image reconstruction, and machine learning.
• 2.3.2022
  RNDr. Michal Zima
  Bonchai: Sustainable bounded-size blockchain
  Abstract: In this talk about blockchains, we will explore some of the dark sides of blockchains. Namely, we will focus on their design flaw: unlimited growth. Why does the design have such a flaw? Or is it a feature? What is behind? Good questions to be answered. To not end there, we will also discuss possible solutions. I will show in detail Bonchai, a solution I work on, and how it can turn a disk-space-hungry 450GiB+ bitcoin blockchain into a 100× smaller one and even guarantee security and a limit on possible growth. That may be significant for embedded devices or SCADA nodes that are rarely upgraded yet need to operate over extended time periods (years or even decades). Does Bonchai come with its own dark sides? Of course, it does! This talk has them covered.
• 9.3.2022
  Ing. Lucie Homolová, Ph.D.
  Remote sensing for monitoring of forest ecosystems
  Abstract: Remote sensing methods allow for spatio-temporal monitoring of landscape changes, functions, structures. In this presentation, we will introduce the activities of the Remote Sensing Department of CzechGlobe. The group operates a rather unique airborne infrastructure operating in the European research space that enables simultaneous acquisition of hyperspectral, thermal, and laser scanning data. We will present how those airborne data, as well as publicly available satellite optical data (e.g., Sentinel-2), are nowadays used for monitoring forest ecosystems.
  PDF
• 16.3.2022
  RNDr. David Střelák
  Acceleration of image processing algorithms for single particle analysis by electron microscopy (rehearsal of Ph.D. thesis defense)
  Abstract: The main goal of the thesis was to improve the algorithms used in Cryo-Electron Microscopy (Cryo-EM), esp. the Single Particle Analysis (SPA). We focused primarily on single-node performance optimizations, using either available or developed techniques in the HPC field, such as heterogeneous computing or autotuning. The secondary goal of the thesis was to identify the limitations of state-of-the-art HPC techniques (by their application in Cryo-EM) and potentially extend those techniques to overcome their limitations. In particular, we focused on the following research topics: reformulation of the existing algorithms with respect to scalability, automatic data optimization, the introduction of the (dynamic) autotuning to the Cryo-EM pipeline, and introduction of the task-based systems to the Cryo-EM pipeline.
  The results presented in the thesis were applied in a real-life, open-source software package XMIPP, which greatly impacted the whole Cryo-EM community. We accelerated several protocols of XMIPP (3D reconstruction, Align Significant, Movie Alignment, and others) typically by over an order of magnitude, compared to the original parallel CPU approach. We have also proposed and evaluated several generic tools for performance optimization, namely cuFFTAdvisor for automatic search of the parameters of the cuFFT library, and contributed to Kernel Tuning Toolkit, a tool allowing for offline and dynamic autotuning of the CUDA and OpenCL kernels. Last but not least, we proposed the Umpalumpa image processing framework, which uses a novel combination of the task-based runtime system and dynamic autotuning and can thus optimize the execution for used hardware and data being processed.
  PDF
• 23.3.2022
  RNDr. Zdeněk Matěj, Ph.D.
  Fast digital technology for nuclear facilities
  Abstract: With the development of digital technologies, there are many possibilities for their application in practice. A combination of fast algorithms together with sufficiently powerful HW based mainly on programmable gate arrays (FPGA) extend existing detection possibilities in the field of nuclear experiments. This talk will describe the development of digital devices that primarily help in the field of detectors and nuclear experiments and also serve for research in the field of IV. generation of nuclear reactors.
• 30.3.2022
  RNDr. Lukáš Ručka, RNDr. Miloš Liška, Ph.D.
  Messing with bare metal in computer networks
  Abstract: In this talk, we introduce the audience to the concept of bare metal and white box switching. The first part of the talk explains how bare metal and white boxes fundamentally differ from traditional networking equipment in both its operations and management and provides bare-metal switches featuring the Intel Tofino chipset as an example of such an approach.
  The second part of the talk introduces the Intel Tofino programmability using the P4 language and some of its applications we are currently working on. One of the goals of this work is to assess the benefits of bare metal equipment and Intel Tofino-based acceleration for the future CESNET backbone. The preliminary results indicate we can successfully saturate a 100G link using ordinary IPv6 ping on even the most trivial topology.
• 6.4.2022
  RNDr. Dalibor Klusáček, Ph.D.
  Kubernetes container orchestrator – Scheduling problems and challenges
  Abstract: This talk will discuss our experience when utilizing the Kubernetes container orchestrator (K8s) to efficiently allocate resources in a heterogeneous and dynamic academic environment. In the commercial world, the „pay per use“ model is a strong regulating factor for efficient resource usage. In the academic environment, resources are usually provided „for free“ to the end-users; thus users often lack a clear motivation to plan their use efficiently. We will discuss how these issues have been resolved in the past (in classic HPC) and identify obvious inefficiencies in existing K8s scheduling abilities. For example, unlike in traditional batch-oriented HPC, users typically require an interactive and waiting-less experience. They do not appreciate waiting, but constantly keeping resources available for interactive tasks is inefficient. The second phenomenon is observable in both interactive and batch workloads; users tend to overestimate necessary limits for their computations, thus wasting resources. Finally, Kubernetes does not support fair-sharing functionality (dynamic user priorities), which hampers efforts to develop a fair scheme for Pod/job scheduling and/or eviction. We discuss various approaches to deal with these problems, such as scavenger jobs, placeholder jobs, Kubernetes-specific resource allocation policies, separate clusters, or priority classes. We also show that all these proposals open interesting scheduling-related questions that are hard to answer with existing Kubernetes tools and policies.
  PDF
• 13.4.2022
  doc. RNDr. Vlastislav Dohnal, Ph.D.
  Metric hull as similarity-aware operator for representing unstructured data
  Abstract: Similarity searching has become widely utilized in many online services processing unstructured and complex data, e.g., Google Images. Metric spaces are often applied to model and organize such data by their mutual similarity. As top-k queries provide only a local view of data, a data analyst must pose multiple requests to observe the entire dataset. Thus, group-by operators for metric data have been proposed. These operators identify groups by respecting a given similarity constraint and produce a set of objects per group. The analyst can then tediously browse these sets directly, but representative members may provide better insight.
  In this talk, we focus on concise representations of metric datasets. We introduce a novel concept of a metric hull that encompasses a given set by selecting a few objects. Testing an object to be part of the set is then made much faster. The metric hulls provide faster and more compact representations when compared with commonly used ball representations. We will show its preliminary applications, e.g., in the direction of indexing metric data.
  video
• 20.4.2022
  doc. Ing. Vojtěch Spiwok, Ph.D.
  Driving molecular simulations by machine learning
  Abstract: Biomolecular simulations have great potential in drug design, protein engineering, and other fields. In principle, it is possible to simulate the binding and unbinding of a drug from its protein target. Similarly, it is possible to simulate the folding and unfolding of proteins. However, such simulations are usually impractical and often impossible due to enormous computational costs. These processes take place in milliseconds or longer time scales. Time steps in biomolecular simulations are very short and typically in the femtosecond range. Thus such simulations require an enormous number of steps. Furthermore, it is necessary to evaluate a huge number of atom-atom interactions in each step. These two factors cause that it is possible to simulate nanoseconds or microseconds routinely. In order to study longer time scales, we use enhanced sampling methods such as metadynamics. Metadynamics disfavors states that are sampled for a too long time. As a result, different slow transitions can be accelerated. The key to the success of metadynamics is the description of the states of the simulated systems. We will present our applications of machine learning methods to describe the progress and accelerate biomolecular simulations.
  video
• 27.4.2022
  Canceled due to CESNET and MetaCentrum seminar
• 4.5.2022
  RNDr. Lukáš Hejtmánek, Ph.D.
  Containers and Kubernetes, or „There and back again“
  Abstract: A short introduction to containers and K8s shall be given in this talk and hands-on. We will discuss the benefits of using containers for research and development. The hands-on part will show how to move from a bunch of scripts and hacks plagued by dependency hell that most users suffer from when trying to use the infrastructure or sometimes even their own desktops to containers. We show how to use or modify existing containers or create specialized containers from scratch. We will also show how to run such containers on users‘ computers and K8s container infrastructure. Examples will be focused on practical problems such as using Caverdock and/or Kernel Tuning Toolkit. The second part of the hands-on will be focused on data handling as many problems with using shared K8s infrastructure are related to accessing data. Therefore, we demonstrate data transfers between a local computer and the infrastructure and how to access data residing on MetaCentrum/CERIT-SC infrastructure, what possibilities the infrastructure offers, what to use, and when.
  Active participation in the hands-on is recommended. For all examples, shell, web browser, and access to docker are required; for some examples, only shell and web browser are needed.
  PDF, web guide, video
• 11.5.2022
  RNDr. Lukáš Hejtmánek, Ph.D.
  Containers and Kubernetes, or „There and back again“ (second part)
  Abstract: Short feedback on the first part will be given in this second talk. As announced, data handling was missed; it will be presented in this part. The demonstration will consist of running an interactive application and transferring data back and forth. A short discussion about other possibilities and connecting to e-Infra resources will follow. The second part of the talk will show a live demonstration of real-case AI computation. We create a new container, upload it to the Docker registry and run it in Kubernetes.
  For the first part of the presentation, a notebook or tablet with a web browser is the only requirement to engage hands-on. For the second, more advanced part, the docker and kubectl tool is required. Docker access shall be provided again via Cerit-SC server, and it will be more convenient if participants can install and configure the kubectl tool as described in the web guide.
• 18.5.2022
  Future Sitola and SitSem planning
• 8.6.2021
  State exam rehearsals (master, bachelor): please be careful, the seminar will last till about 16:30
  Václav Sobotka: Vehicle routing with transfers
  Supervisor: Hana Rudová
  Reader: David Woller

  Samuel Gorta: Exponential mathematical models of MRI data
  Supervisor: Aleš Křenek
  Reader: Tomáš Brázdil

  Boris Jurič: Implementation of PDX model sequencing data analysis in Galaxy
  Supervisor: Aleš Křenek
  Reader: Radka Svobodová

• 15.6.2021 State exam rehearsals (bachelor)
  Tereza Takáčová: License Usage Monitoring
  Supervisor: Eva Hladká
  Reader: Dalibor Klusáček
• 22.6.2021
  State exam rehearsals (bachelor)
  Matej Hrica: Screen capture for UltraGrid (wayland + X11)
  Supervisor: Lukáš Ručka
  Reader: Martin Pulec

  Kryštof-Mikuláš Štys: Platform for automation devices using ASP.NET 5, REST and MQTT
  Supervisor: Eva Hladká
  Reader: Leonard Walletzký

  Matěj Bukáček: Comparison of cortical layer signal reconstruction methods in magnetic resonance data
  Supervisor: Aleš Křenek
  Reader: Jan Fousek

• 8. – 10.9.2022 (based on 3 day plan)
  SitSem at Telč

Contact: Hana Rudová