Autumn 2021 Program

This is the programme of the Sitola seminar in autumn 2021. Presentations for the current semester are available here.

• 15.9.2021
  Mgr. Jiří Vorel
  How to handle MetaCentrum in a few steps
  Abstract: This lecture is dedicated to the beginning users of MetaCentrum. We will talk about how to reserve computational resources and prepare a batch or interactive job properly. We will continue by describing available queues, showing how to handle big data and what mistakes users should avoid.
  PDF, video
• 22.9.2021
  Mgr. Aleš Křenek, Ph.D.
  Calibrating machine-learning models to increase the prediction accuracy of specific experiments
  Abstract: The possibility to predict an outcome of some future process has virtually always certain values, should it be a laboratory measurement, weather forecast, or parliamentary election. Numerous machine learning techniques were developed to make such predictions based on knowledge of past outcomes of such processes. The common requirement of these approaches is the availability of a sufficiently large and internally consistent data set of such past outcomes.
  In our specific case, the liquid chromatography, meeting this requirement is particularly challenging for two reasons — setup of every laboratory and equipment is quite unique, and the results are not directly transferable among the labs, and the variability of analyzed samples is huge.
  Recent results show that the problem can be mitigated by combining multiple machine learning models trained on several large reference datasets (past results of well-established large laboratories) with measurements of small size, affordable *calibration* experiment done in the target laboratory setup. The reference datasets capture the complexity of the model while the calibration anchors the prediction to the specific setup.
  The talk is heavily based on the results published in https://doi.org/10.1021/acs.analchem.0c00233, which we successfully applied in the Recetox laboratory at MU.
  video
• 29.9.2021
  Mgr. Zuzana Müllerová, M.Sc.
  Practice of course timetabling with UniTime at Masaryk University
  Abstract: UniTime is a comprehensive open-source educational scheduling system. It has been used for course timetabling at Masaryk University for the past ten years. Seven faculties have gradually decided to benefit from it. Masaryk University has been involved in its research and development. In this talk, the course timetabling problems at the Faculty of Informatics and six other Masaryk University faculties will be presented, emphasizing the differences between them. We will also cover interesting aspects of course timetabling with UniTime at institutions from other continents if time allows.
  PDF
• 6.10.2021
  Ing. Jan Hoidekr, Ph.D.
  Running containers in Metacentrum
  Abstract: Containers with singularity extend the capabilities of the Metacentrum for tasks with many library dependencies and support programs in the fields such as AI, Machine/Deep Learning, and GPU jobs, e.g., PyTorch or Alphafold2. Will introduce the concept of containers and show the principles which offer easy repeatability of calculation and sharing of the computing environment.
  PDF
• 13.10.2021
  RNDr. Matej Antol
  Advancing management of complex data in metric spaces (rehearsal of Ph.D. thesis defense)
  Abstract: Most of the data emerging in modern applications are complex, which presents specific challenges in their processing. To manage, retrieve and analyze complex data, the similarity between data objects is modeled as their pairwise distance. These distances are then used for complex data management based on the mathematical concept of metric spaces.
  This talk will discuss original approaches to partitioning, indexation, and similarity searching of complex data using the metric space model. To describe data clusters, we will describe a brand new algorithm that can create compact representations of metric data similar to convex hulls. In indexing, the talk will present a modification of one of the currently-used indexes, which allows it to distribute data evenly between data partitions. Improvements to searching are introduced by considering information about the historical popularity of data and their clusters – exploiting this property, as well as any other data properties such as the sizes of clusters, serves to create more holistic and better-performing priority queues. The talk also introduces learned indexing to the realm of metric data management, which challenges the standard paradigms of (complex) data indexation.
• 20.10.2021
  Ing. Václav Oujezský, Ph.D.
  Network programmability and automation
  Abstract: Why do we need „Network Programmability or Automation“? Given the scale and complexity of today’s computer networks, it is impossible to ensure their proper operation without centralized management. Be managing a network with 200 or more network devices and more than 1,000 different running configurations; we need to respond flexibly to all requests and changes in a reasonably short time. For example, it can be a change of authentication on many network devices, deployment of a new system, virtualized operating systems, or the resolution of incidents and security events. With increasing demands on quality and cost reduction, it is practically impossible to achieve this without systemic change in the classic networking concept. This change is brought about by Software-Defined Networks and network programmability. The practical demonstration will show what is meant by programmability and automation, also what tools or programming (scripting) languages are used in practice.
• 27.10.2021
  Mgr. Filip Petrovič
  Visual programming language for open-world RPG quest development
  Abstract: In recent years, visual programming languages have become commonly used in many areas of digital game development, ranging from AI to audio programming. We will have a look at one such language, which is used specifically for implementing quests in the open-world RPG game developed at one of the Czech studios. The lecture will involve a high-level overview from the user perspective and then dive into interesting implementation details.
• 3.11.2021
  doc. RNDr. Barbora Kozlíková, Ph.D.
  Visual analysis of long molecular dynamics simulations
  Abstract: One of our long-term research efforts is to support the biochemist within the exploration of very long simulations of molecular movements (molecular dynamics), where they want to observe interesting events occurring over time. To support the detection and visual exploration of such events, we designed an interactive focus+context visualization that helps the biochemists in navigation through the spatial and temporal complexity of the data. In this talk, I will present our approach and outline our ongoing research efforts in this direction.
• 10.11.2021
  Bc. Martin Kurečka, Mgr. et Mgr. Jaroslav Oľha
  Performance issues in molecular simulation software: a failure story, a success story
  Abstract: Most of the modern scientific programs designed for complex simulations have two things in common. Firstly, they are not made by professional expert programmers, and secondly, they are large, complicated, and poorly documented as a result of years (or even decades) of incremental feature creep. We have spent several years working with one such program, a molecular dynamics library called PLUMED, all the while attempting to pinpoint and fix some of its performance issues. This is a story of our failures and eventual partial success.
• 24.11.2021
  RNDr. Vít Rusňák, Ph.D.
  Zettelkasten: A method for personal knowledge management
  Abstract: Whether you are a student, teacher, or researcher, information processing is a core skill that you use daily. The process of note-taking and its transformation into knowledge that we can use in our work is known as Personal Knowledge Management (PKM). As a researcher, I read and make notes a lot. However, I realized that my system was far from optimal. Earlier this year, I decided to change it. Zettelkasten, the method I am adopting, was developed by German social scientist Niklas Luhmann in the 1970s. The technique has recently been re-discovered and popularized by Sönke Ahrens in the book „How to Take Smart Notes“. Although I am still learning the process, I am more and more convinced that having some form of PKM is becoming a must for all the knowledge workers.
  The talk aims to introduce the concept of PKM, show you the importance and impact of note-taking and information processing for your future selves, and (perhaps) motivate you to start your own. We will discuss your note-taking approaches and whether you have some experience with PKM systems or not. Next, I will introduce the Zettelkasten method and demonstrate my current (and planned) setup. I will also give you some pointers to other PKM methods and tools you can try.
  PDF
• 1.12.2021
  Stephen Tsirtsonis
  Troubleshooting and security response in 100 Gig environments
  Abstract: Endace has been at the high end of network communication monitoring for the past 20 years and developed the first 100Gig solution back in 2012. In this presentation, we will talk about what is required in order to understand what happens over these critical core connections to counter both cyber threats and performance issues, as well as how to integrate such a system into existing tools to simplify and accelerate incident response.
• 8.12.2021
  Mgr. Petra Němcová
  Improving ligand transport trajectory within flexible receptor in CaverDock (rehearsal of the defense proposal for Ph.D. thesis)
  Abstract: The receptor-ligand interactions are an important part of many biologically relevant processes. The small ligand molecule needs to pass via a tunnel into a receptor before the interaction of interest begins. Both receptor-ligand interaction and ligand pathway need to be studied. CaverDock is a computational tool that simulates the transport of the ligand in a tunnel. However, in its first version, CaverDock allowed only limited flexibility of the receptor, biasing the measured energy of ligand transportation. We combine its force field with AMBER, which allows for relaxing the receptor’s geometry in tunnel bottlenecks. Results are demonstrated on the experimental data set.
• 15.12.2021
  Christmas meeting outside
• Examination period 3.1. – 11.2.2022
  Meetings devoted to state exam rehearsals are scheduled mostly. State exam rehearsals are in Czech mostly.

• 26.1.2022
  State exam rehearsals (master and bachelor)
  Metoděj Klang: Dynamic vehicle routing with rich problem characteristics
  Vedoucí: Hana Rudová
  Oponent: Dalibor Klusáček
• 2.2.2022
  State exam rehearsals (bachelor)
  Michal Starý: Prediction of missing peaks in mass spectra
  Vedoucí: Aleš Křenek
  Oponent: Vít Nováček
• 9.2.2022
  RNDr. Tomáš Raček
  Empirical methods for calculating partial atomic charges (rehearsal of Ph.D. thesis defense)
  Abstract: Partial atomic charges are a theoretical concept that simplifies the description of the electron density in a molecule. They find many applications in fields such as bioinformatics, biochemistry, or cheminformatics. Partial atomic charges cannot be obtained from experiments; their values must be determined by calculation. The most common method is the use of quantum mechanics (QM), but the computational complexity of these methods limits their application to very small systems. In the last more than 50 years, however, a large number of so-called empirical methods for calculating partial atomic charges have been developed. This group of methods uses parameterized or experimental data. As a consequence, they are significantly faster than QM methods and thus applicable even for large biomolecules such as proteins. This talk addresses several topics in the field of empirical methods. First, we will show how existing methods can be improved, both manually for specific groups of molecules and semi-automatically using symbolic regression and genetic programming in a domain-specific language we have developed. Next, we will focus on the parameterization process, which allows us to tune these empirical methods to reproduce a particular type of QM charge with high accuracy. Finally, we will talk about making the results accessible to the scientific community.