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 audience

  • 21.2.2024
    Student presentations with ongoing work (organized by Václav Oujezský)

    Adriana Fojtíková
    Digitization of educational courses
    Annotation: The Learning Management System (LMS) is a tool that makes education more accessible and engaging for students. This bachelor thesis focuses on comparing multiple systems, and in the practical part, educational materials from the courses PV169 and PA151 will be deployed on this platform.

    Filip Fabka
    Visualization of data-link control protocols
    Annotation: Data-link Control Protocols are used to regulate the transmission and reception of data in the data link layer (L2). The goal of my bachelor thesis is to visualize these protocols in a web environment for subsequent use in educational materials for the subject PV169.

    Bc. Michal Badin (presented by doc. Ing. Václav Oujezský, Ph.D.)
    Development of a federated learning platform with firebase database integration
    Annotation: This master’s thesis deals with the implementation of a federated learning platform, specifically a server with an admin dashboard in a web environment, building on the work – A proposal for a Federated Learning Protocol for Mobile and Management System. The objective of the thesis is to address the problems of a centralized server architecture and enable smooth collaboration with multiple centralized servers, with a focus on user data privacy. This is achieved by ensuring that sensitive data does not propagate when communicating data to the server for model training.

  • 28.2.2024
    Ing. Klára Moravcová, CESNET
    OpenStack under the hood: new architecture
    Abstract: If you have ever dabbled in OpenStack, creating virtual infrastructures or running applications, you might have pondered what it truly entails to operate such a platform. Today, we are diving deep into the inner workings of OpenStack, uncovering the essential components needed for smooth operation. This presentation offers an in-depth look into the transformation of our OpenStack architecture into Beskar Cloud—a cutting-edge architecture that makes use of modern automation tools and follows the GitOps approach.

  • 6.3.2024
    Student presentations with ongoing work (organized by Hana Rudová)
    Notino company has a complex large warehouse that consists of several floors. Tens of thousands of orders are handled daily in the warehouse, each consisting of several items. A moving conveyor belt with boxes passes through each floor of the warehouse. Each box collects the goods for one order at a time. It is necessary to assign the floor from which each box is served, and several pickers collect the items for the orders on each floor. Overall, it is necessary to maximize warehouse throughput and ensure that orders are processed in a timely manner.
    This seminar presents a series of ongoing theses that deal with warehouse planning for Notino. We introduce a warehouse simulator, a planner that focuses on distributing orders to individual floors, and a planner that focuses on optimizing the operation of a single floor.

    Bc. Milan Kubík
    Notino warehouse simulator

    Bc. Samuel Krivánek
    Optimization of warehouse throughput for the Notino company

    Bc. Martin Blažek
    Order picker routing in the warehouse of Notino company

  • 13.3.2024
    Ing. Jan Král, Ph.D., FI MU
    Digital predistortion in wireless communication systems
    Abstract: Wireless communication plays a vital role in our daily lives. Digital predistortion (DPD) is a key technology that improves the quality of transmitted signals, enabling higher transmission throughput and reliability of wireless networks. To meet today’s speed and throughput demands, efficient DPD implementations are essential. In this lecture, I will present solutions for efficient DPD implementations.
    First, I will introduce the basic principles of digital predistortion and explain how this technique is employed to linearize radio-frequency power amplifiers. I will then present novel advanced digital signal-processing techniques for decreasing the computational complexity of the DPD adaptation. I will particularly focus on an approach to selecting proper feedback samples, which can reduce the computational complexity of the adaptation by up to 400 times. Subsequently, I will discuss effective implementations of DPDs in modern digital platforms, for example via lookup tables in Field Programmable Gate Arrays. I will conclude the talk by laying out my research vision for the integration of machine learning techniques in the domain of digital predistortion.

  • 20.3.2024
    Mgr. Tomáš Foltýnek, Ph.D., FI MU
    Ethical challenges of artificial intelligence
    Abstract: Recent advancements in large language models have captivated the general public with the power of AI systems, prompting questions about the benefits of these technologies for human society and the limits of capabilities we ought to pursue. This lecture aims to provide a comprehensive overview of the ethical challenges posed by AI, targeting computer science researchers from diverse fields. We begin by addressing foundational issues of bias and fairness in algorithmic decision-making, emphasizing the importance of developing transparent, equitable AI systems.
    We then delve into the impact of AI on the workforce, discussing ethical considerations around automation and its effects on employment. A significant portion of our discussion will focus on the integration of AI in education, where we will examine both the potential benefits and the ethical dilemmas it presents, including personalized learning, data privacy, and the digital divide.
    Furthermore, the lecture addresses the crucial aspects of responsibility and accountability in AI deployment. We will explore scenarios of unintended consequences, such as the proliferation of fake news, and discuss the role of developers and the AI community in ensuring ethical compliance. Amid these challenges, the need for robust ethical frameworks and guidelines to govern AI research and deployment becomes evident. We will review existing initiatives and propose directions for future research, with the goal of fostering an environment where AI contributes positively to society while minimizing harm.

  • 27.3.2024
    RNDr. Jiří Filipovič, Ph.D., ÚVT MU
    What can we expect from code performance optimization? (presentation for the habilitation defense)
    Abstract: In this talk, I will demonstrate the importance of code transformations focusing on decreasing its runtime by better hardware utilization. We start with the naive C implementation of an algorithm used in scientific computing, which does not take into account the properties of modern hardware. Then, I introduce some features of modern processors (caches, vector instructions, limitations of multithreading) and make changes in the code to reflect those features. Finally, I will show how to transform the code to GPUs and get even more performance, presenting our work pushing the state-of-the-art.

  • 3.4.2024
    doc. RNDr. Jan Sedmidubský, Ph.D., FI MU
    Text-to-motion retrieval: Towards joint understanding of human motion data and natural Language
    Abstract: Due to recent advances in pose-estimation methods, human motion can be extracted from a common video in the form of 3D skeleton sequences. Despite wonderful application opportunities, effective and efficient content-based access to large volumes of such spatio-temporal skeleton data still remains a challenging problem. Inspired by the recent progress in text-to-image/video matching, we propose a novel content-based text-to-motion retrieval task, which aims at retrieving relevant motions based on a specified natural-language textual description. To define baselines for this uncharted task, we employ the BERT and CLIP language representations to encode the text modality and successful spatio-temporal models to encode the motion modality. We set up a common evaluation protocol by defining qualitative metrics for assessing the quality of the retrieved motions, targeting two datasets: KIT Motion-Language and HumanML3D. This work was awarded the Best Short Paper Award Honorable Mention at the SIGIR 2023 conference (Core A*).

  • 10.4.2024
    Mgr. Viktória Spišaková, FI MU
    In the container era: A coup in reliable computing over unreliable infrastructure (presentation for the thesis proposal defense)
    Abstract: The contemporary computational landscape is dominated by virtualization and cloud computing, marked by the utilization of virtual machines and modern containers. Modern clouds promise infinite scaling, paying only for what was used, several nines of reliability, and many other perks. However, such grandiose pledges come with a toll – inefficient resource utilization – because of the need to provision for the peak load and reservations. Suboptimal resource utilization has been a persistent topic in the computer science community since the 60’s, and many scheduling techniques have been developed to improve resource allocations. Although the ultimate scheduling oracle is impossible, revisiting the already-known concept of backfill/fill-in/scavenger workloads may lead to enhanced resource allocation in the cloud environment with the current conceptual and technological state-of-the-art – containers and their lightweight checkpoint/restore mechanism. There are only a few workload types that are a natural fit for scavenger workloads, so we propose to enlarge a pool of possible workloads that may run in the background by making every workload interruptible. However, sudden interruption introduces unreliabilities into the system, such as no guarantee that the scavenger workload ever finish or that the scavenger workload will not waste the resources again. Therefore, a sophisticated framework using container checkpoint/restore must be created to address these unreliabilities while maintaining the cloud’s elasticity property. This thesis proposal researches possible tools that can facilitate the development of such a framework, introduces the design of a novel scheduling technique that optimizes according to the computed CPU time, thus reducing resource utilization inefficiencies, and explores ways of performing the mentioned actions in a distributed manner.

  • 17.4.2024
    Canceled due to CESNET seminar

  • 24.4.2024
    Student presentations with ongoing work (organized by Tomáš Rebok)

    Pavel Kraus
    Language models and their use for the needs of security forces
    Annotation: The lecture will focus on the presentation of the currently ongoing application-research activities related to the examination of available language models and frameworks supporting interaction in the Czech language. The presentation will also address the considered application-research scenarios and applications to different areas of work of security forces, especially the Police of the Czech Republic.

    Jakub Drábik
    NERvana – semantic information mining from heterogeneous datasets
    Annotation: The presentation will present the developed NERvana solution for the needs of security forces enabling (semantic) entity mining in large heterogeneous data sets – especially in electronic documents of various formats. It will briefly present both the integrated technologies enabling semantic identification of entities of interest (names, emails, addresses, phone numbers, etc.), as well as the experience from deploying this solution in the environment of the Police of the Czech Republic and directions for future research.

    Michal Klinka
    Data-analysis infrastructure „on-demand“
    Annotation: This presentation will present the developed solution of flexible and modular data-analytical infrastructure applied to the needs of AFoLab, EnviLab and eLTER projects. The presented solution enables containerized data processing and analysis in the form of pipelines (workflows) performing custom computations on top of a Kubernetes cluster. In addition to a brief introduction of the integrated technologies and capabilities of the infrastructure, the presentation will mainly focus on the automated deployment and mass management of its instances, employing the so-called GitOps principles.

  • 15.5.2024
    Seminar about Sitola and SitSem organization

Past seminars

Contact: Hana Rudová

(c) 2021 SITOLA, Administration