![\"Our](\"https:\/\/www.sitola.cz\/wordpress\/wp-content\/uploads\/2018\/04\/seminar-180418-long-s-3.png\")
<\/p>\n<\/p>\n
\nLaboratory seminars (PV273 in the course catalog) on Wednesday, 10:30 – 11:30<\/b>, A505, FI MU, Botanick\u00e1 68, followed by an informal group lunch.<\/p>\n
The format of standard lectures: 30-40 minutes presentation + 15 minutes for questions. Slides are expected in English, and the presentation is in English or Czech, depending on the audience. For past seminars, go here.<\/a><\/p>\n\n
\nRNDr. Luk\u00e1\u0161 Hejtm\u00e1nek, Ph.D.
\nThe unbearable lightness of AI<\/strong>
\nAbstract: The CERIT-SC Center launched its AI large language model (LLM) inference service one year ago now. What started with comparatively small models\u2014around 70B parameters and roughly 40GB of weights\u2014has since evolved into a large-scale production environment. Today, we operate several models approaching 700B parameters and one 1000B-parameter model, all running on enterprise-grade DGX B200 and B300 hardware. This presentation highlights recent advances in AI with a particular focus on operating LLM inference services at CERIT-SC. We share practical experience gained over the past months, including architectural decisions, operational challenges, and key lessons learned while scaling from early deployments to state-of-the-art infrastructure. Beyond inference itself, we present the broader ecosystem of AI-enabled services built around these models. This includes agentic systems such as n8n, integrations with developer tools like VS Code, terminal environments, and Jupyter Notebooks, as well as the MCP servers we operate. We explain how these components fit together, why they form a critical part of the overall system, and how they support user-facing services such as chat.ai.e-infra.cz. Finally, we demonstrate how this AI ecosystem translates into real productivity gains, accelerating everyday workflows and reshaping how we approach research, development, and operational tasks.<\/li>\n
\ndoc. Mgr. Pavel Rychl\u00fd, Ph.D.
\nUnderstanding LLMs<\/strong>
\nAbstract: Large Language Models (LLMs) have generated significant excitement and high expectations, including aspirations for Artificial General Intelligence, yet their inner workings are often misunderstood.
\nIn this presentation, we will demystify the magic behind LLMs\u2014from the challenges of language itself to the power of word embeddings and transformers. We will show how GPT models process words, capture context, and create meaningful responses. We will explore the role of attention mechanisms, the flow of information through neural networks, and why these technologies are reshaping the future of AI.<\/li>\n
\nRNDr. V\u00e1clav Sobotka
\nKey principles in cross-domain hyper-heuristics<\/strong>
\nAbstract: Cross-domain selection hyper-heuristics typically focus on adaptively selecting search operations, so-called low-level heuristics (LLHs), from a predefined set. In contrast, we concentrate on the composition of this set and its strategic transformations. We systematically analyze transformations based on three key principles: solution acceptance, LLH repetitions, and perturbation intensity, i.e., the proportion of a solution affected by a perturbative LLH. We demonstrate the raw effects of our transformations on a trivial, unbiased random selection mechanism. Additionally, we accompany several recent hyper-heuristics with such strategic transformations, often effectively simplifying their designs. Using this approach, we show the three aforementioned principles as simple yet powerful drivers of cross-domain search performance and outperform the current state-of-the-art hyper-heuristics on both the standard CHeSC cross-domain benchmark and three real-world domains.<\/li>\n
\nMgr. Samuel Gorta
\nExtending molecular dynamics simulations via latent space projection<\/strong>
\nAbstract: Molecular dynamics (MD) simulations provide critical insights into the microscopic behavior of systems comprising thousands to millions of atoms. However, the high computational cost associated with these simulations severely restricts the accessible trajectory lengths, effectively limiting the range of observable physical phenomena.
\nRecent methodological advancements (e.g., https:\/\/doi.org\/10.1039\/D0SC03635H<\/a>) demonstrate that the complex dynamics of such systems can be effectively approximated within a reduced-dimensional latent space. This approach is rooted in the theory of transfer operators, which model the temporal evolution of probability distributions across the system’s state space. The latent space is formally defined by the basis of the transfer operator\u2019s eigenfunctions. By ranking these basis functions according to their corresponding eigenvalues, less significant dynamical components can be truncated, achieving substantial dimensionality reduction while preserving the essential kinetics of the system. In this framework, the embedding of a specific molecular configuration into the latent space represents its probability values across the selected basis functions.
\nCritically, this mapping can be modeled using machine learning techniques without the need for an explicit analytical expression of the basis. By training on relatively short input trajectories, these models can learn the latent dynamics, offering a promising path toward extending the temporal reach of MD simulations far beyond current computational limits.<\/li>\n
\nRNDr. Tom\u00e1\u0161 Rebok, Ph.D.
\nBuilding the eLTER cyberInfrastructure: From H2020 development to operational data services for a European research infrastructure<\/strong>
\nAbstract: The eLTER CyberInfrastructure, which we have developed through our participation in the eLTER PPP and eLTER PLUS H2020 projects, illustrates how modern IT can support large-scale, distributed environmental research infrastructure across Europe. This talk presents the modular eLTER Cyberinfrastructure solution, which integrates data registration, metadata harmonization, quality-control workflows, analytical environments, and other user-facing services into a single coherent and interoperable solution. The talk highlights the design and architecture principles behind such a sustainable research infrastructure, including cloud-native, automated deployment and management, workflow-based data processing, and FAIR and interoperable data management. Additionally, we will discuss the challenges of transitioning from project-driven development to sustainable operational services within a European research infrastructure, and share lessons learned from our active participation in a diverse research community. The presentation will also introduce the emerging eLTER ERIC and the services of one of its core facilities\u2014the eLTER ERIC Topic Centre for Data Management\u2014which we recently successfully applied to host.<\/li>\n
\ndoc. RNDr. Radka Svobodov\u00e1, Ph.D.
\nWhere are electrons in molecules? Mapping probabilities to real numbers<\/strong><\/strong>
\nAbstract: While everyday objects – like a chair or a laptop – can be easily located in space, it is not that simple for microscopic particles like electrons and photons. In this micro-world, we only know the probability of finding them in a certain area.
\nYet, understanding the spatial distribution of electrons, commonly referred to as electron density, is the key to deciphering a molecule’s chemical behavior. To bridge this gap, we must map the probabilistic distribution of electrons onto real numbers that approximate the electron density assigned to individual atoms. These values are known as partial atomic charges. Whether derived from rigorous quantum mechanical calculations or through various approximations and heuristics, calculating these charges is a fundamental challenge. This lecture will delve deeper into the intricacies of this fascinating concept.<\/li>\n
\n\u0160tefan Morav\u00edk
\nDesign and implementation of a drone mission planning module for airport lighting inspection<\/strong>
\nAnnotation: This thesis develops a software module for automated drone mission planning to inspect airport Precision Approach Path Indicator lighting systems. The solution generates safe, collision-free flight paths within active airport environments and provides a web-based interface for trajectory visualization and mission export, aiming to replace costly manned flight inspections with objective, repeatable UAV-based measurements.
\nSamrawit Fentaye Abebe
\nOptimizing software development and deployment with multi-agent Large Language Models<\/strong>
\nAnnotation: This talk presents a multi-agent approach to automate the software development lifecycle process using Large Language Models. I will discuss the system architecture, orchestration strategies, and safety constraints, with a focus on controllability, auditability, and applicability in the enterprise context.
\nBc. Tom\u00e1\u0161 Men\u0161\u00edk
\nMulti-agent systems for cybersecurity event detection in computer networks<\/strong>
\nAnnotation: Current cybersecurity tools detect threats well but operate in isolation, leaving data fragmented. We test whether a multi-agent system using LLMs can bridge this gap by having autonomous agents collaborate to link scattered signals better than traditional rule-based methods. We compare three leading frameworks, implement a prototype on the vSafe backup platform with specialized agents, and evaluate the results on public and internal datasets. This talk outlines the architecture, interaction models, and validation metrics to determine if agent-based reasoning can effectively restore missing context and turn raw alerts into actionable decisions.<\/li>\n
\nAndrej Tejbus
\nMerge conflict synthesis in an existing Git project<\/strong>
\nAnnotation: This thesis focuses on the design and implementation of a new laboratory in the PB176 course. The primary goal is to create an environment for practicing the resolution of merge conflicts. Students are able to use their own existing git based projects as a baseline, which they reimplement. During this process, merge conflicts are generated based on the differences between the original and the newly developed code.
\nJakub Skopal
\nAutomatic detection of common mistakes in bug reports<\/strong>
\nAnnotation: This thesis introduces an automated Tier 1 feedback system designed to validate student bug reports in the PB176 course, which allows for a more effective resource allocation by eliminating manual iterations.
\nAdam Lukavec
\nDesign and implementation of an automated environment for networking courses <\/strong>
\nAnnotation: The talk presents the design and implementation of a service for managing virtualized resources and their access in a networking laboratory built on Stratus.FI infrastructure. The service supports user authentication via faculty xlogin, authorization based on course enrollment through the IS API, and integrates with Ansible for automated resource provisioning. I will discuss the overall architecture and the current state of the implementation.<\/li>\n
\nCanceled due to the CESNET seminar<\/li>\n
\nMgr. Ter\u00e9zia Fialov\u00e1, Ing. David Tampier a Tom\u00e1\u0161 \u0160atura, MEng.
\nShort-term PV forecasting based on Total Sky Imager<\/strong>
\nAbstract: Accurate forecasting of photo-voltaic (PV) generation has the potential to significantly improve the economy of PV installations and to contribute towards better grid stability. As PV generation is notorious for fluctuations due to the transition cloud conditions, it is of interest to introduce a tool for predicting irradiance that is observed precisely at the location of a PV site. This work leverages Total Sky Imager and uses ConvLSTM to forecast mean irradiance at the given location for the following ten 5-minute windows. The irradiance is then utilized to compute PV output, with indications that such an approach outperforms the accuracy of forecasts based on satellite imaging.<\/li>\n
\nRNDr. Pavel Troubil, Ph.D.
\nTBA<\/li>\n
\nRNDr. Ter\u00e9zia Slanin\u00e1kov\u00e1, Jakub \u010cill\u00edk
\nLessons learned in building a search engine for 1 billion protein structures<\/strong>
\nAbstract: TBA<\/li>\n
\nSeminar about the organization of SitSem 2026 <\/strong>(10-12.9.2026)<\/li>\n<\/ul>\n