Sitola » Seminar

Seminar

Our seminar

Laboratory seminars (PV273 in the course catalog) on Wednesday, 10:30 – 11:30, A505, FI MU, Botanická 68, followed by an informal group lunch

The format of standard lectures: 30-40 minutes presentation + 15 minutes for questions, slides in English, presentation in English or Czech, based on the audience

19.2. 2025
Ing. Eva Výtvarová, Ph.D.
Brain functional connectivity patterns
Abstract: In this talk, I will introduce the concept of functional connectivity as a tool to map patterns of synchronized activity measured over a period of time across the human brain. Emphasizing underlying mathematics, I will present our work on identifying spatial, temporal, and frequency-specific patterns. Explicitly, these methods will be employed and discussed: principal component analysis, eigenvector centrality, Hidden Markov models, and fluidity.
26.2. 2025
RNDr. Vít Musil, Ph.D.
Optimization layers in neural networks
Abstract: While neural networks excel at learning complex patterns from raw data, they often struggle with tasks that involve combinatorial or algorithmic reasoning. In contrast, graph algorithms and classical optimization methods offer strong solutions for such problems when given clean specifications. This talk explores how these two areas can be effectively combined. We provide an overview of methods that enable end-to-end training of compositional architectures by incorporating (discrete) optimization problems into neural networks to enhance their algorithmic capabilities.
5.3. 2025
doc. Ing. Václav Oujezský, Ph.D.
Innovations in teaching – internal support for teaching quality
Abstract: The presentation will introduce the program of support for teaching innovations in courses of Masaryk University degree programs. It will provide insight into the PV169 Innovation Project with practical examples from project submission to implementation in the context of sharing good practices.
12.3. 2025 14:00
Daniel Ricardo Echeverri Giraldo, Ph.D.
Beyond bullet points: Designing effective presentations
Abstract: This workshop/presentation provides a practical introduction to graphic design principles for creating effective presentation slides. Participants will learn key techniques for visual communication, layout, typography, and image use to elevate their presentations and captivate their audience.
You can bring some of your presentations, and we talk about them!
See presentation.
19.3. 2025
Mgr. Monika Čechová, Ph.D.
The bioinformatics of complete human genomes
Abstract: We are now finally able to sequence complete human genomes using a combination of the newest assembly algorithms and sequencing technologies, each with distinct error profiles and biases. This allows us to study previously inaccessible parts of human DNA, such as centromeres or telomeres. Finally, the soon-to-be availability of hundreds of human genomes brings challenges in how to represent them and how to store and query their genetic variation.
26.3. 2025
doc. RNDr. Jiří Filipovič, Ph.D.
Overview of auto-tuning research in Sitola
Abstract: Since 2017, the High-Performance Computing Research Group has focused on a method enabling automatic application adaptability for maximizing its performance for given hardware and input: autotuning. We develop our own autotuning framework, Kernel Tuning Toolkit, and use it as a cornerstone for extending the state-of-the-art in autotuning. In this talk, I introduce autotuning as a concept that enables performance portability, and I will also introduce our group’s results and future research directions.
2.4. 2025
Branislav Bubán
Fine-tuning Large Language Models on private data
Abstract: Large Language Models (LLMs) have demonstrated remarkable capabilities across various domains, but their general-purpose training often lacks the specificity needed for specialized tasks. In comes the saviour Fine-tuning, but it has a twist. Companies that work on new LLMs need a lot of data to train them on. Therefore if you want to fine-tune a model on private data you can’t rely on online trainers to do the work for you. This presentation is about my work at CESNET creating a chatbot for UltraGrig, how it turned into my bachelor’s thesis and how you can use this work to train your own LLMs.Adriana Fojtíková
Git commit history verification for the PB176 course
Abstract: Gitbot is an automated tool that simulates Git-related scenarios for the PB176 course. One of its main tasks is to check whether the submitted Git history follows the expected commit relationships.
The presentation will introduce my bachelor thesis, which focuses on enhancing the existing Gitbot implementation by utilizing directed graphs to represent Git history, where commits are nodes and edges represent relationships between these commits. By defining rules for evaluating Git history, a rules graph will be compared with the student’s history graph. Based on this comparison, Gitbot will assess whether the student has successfully completed the task. This new approach enables more accurate and efficient evaluation of Git histories, improving automated evaluation within the Gitbot environment.
9.4.2025
Ing. Eva Výtvarová, Ph.D.
Complex networks: IV124 crashcourse introduction
Abstract: I will introduce the IV124 Complex Networks course in this talk. Within 30 to 40 minutes, I would try to convince you that networks are all around us and that you miss a lot by not seeing them. Hopefully, at the end of this talk, you will all be complex networks enthusiasts and will enroll in the course next year.
16.4. 2025
Mgr. Eva Maršálková, RNDr. Terézia Slanináková
Refining structure-based search in AlphaFold DB: Including AlphaFold’s prediction confidence measures into AlphaFind
Abstract: AlphaFind is a tool for structure-based similarity search in the AlphaFold database of predicted protein structures. However, one of its current limitations is that not all parts of a protein structure are predicted with the same confidence, and the search does not account for this variability. In this talk, we present strategies to integrate prediction uncertainty metrics into the search process, allowing AlphaFind to prioritize highly confident structural regions and deliver more accurate and reliable similarity results.
23.4. 2025
canceled due to CESNET seminar
30.4. 2025
Next SitSem
The next SitSem seminar take place in Telč again. Save the dates September 11-13, 2025. And come to discuss what you would like to have there!
7.5. 2025
Mgr. Adrián Rošinec
Molecular dynamics data in the exascale era
Would it not be useful to quickly retrieve and share information about dynamical properties, conformations, and chemical characteristics of a protein or determine whether a structure has been part of existing MD simulations? That is the aim of my PhD project – to design a database and tools for MD simulations, ensuring proper annotation and integration with existing resources such as PDB. This includes developing tools to automatically annotate the data, determine the structure, and, based on similarity search, enable highly efficient searches within this large and distributed world of MD data.
14.5. 2025
RNDr. Miloš Liška, Ph.D.
GPUJPEG goes to space
Abstract: GPUJPEG is a high-performance GPU accelerated JPEG codec developed in Sitola. We have recently been tasked with transcoding European Space Agency (ESA) Copernicus initiative observation data from Sentinel-1 and Sentinel-2 missions. While this initially appeared to be an easy task, we have hit a number of obstacles along the way. This presentation won’t go into gory technical details of the GPU JPEG encoding but shows how an initial simple task may quickly become a deep rabbit hole, what to expect of data stored in real repositories, and how current scientific infrastructures may still not be ready for interactive tasks over datasets with size in order of meager gigabyte

Examination period

28.5.2025
State exam rehearsals (master’s)
TBA
4.6.2025
State exam rehearsals (master’s and bachelor’s)
TBA
11.6.2025
State exam rehearsals (bachelor’s)
TBA

Past seminars

Contact: Hana Rudová