Since 10 November our faculty has a new website! The old website will still be available at for now. Something is broken? Please report it to or use our webform.

Once a week during the academic year, an invited speaker (from abroad, as well as from the Czech Republic) talks about his or her scientific work. Colloquium takes place at the Faculty of Informatics and is open to the scientific community. Lecture dates can be found in the programme. Tuesday 14.00 - 15.00, D2, FI MU, Botanická 68a

Colloquia programme with abstracts for the Autumn 2017 semester

3. 10. 2017
Mgr. Jan Obdržálek, Ph.D., FI MU
Digraph Width Measures
Abstract: Treewidth, defined by Robertson and Seymour, proved to be an extremely successful graph parameter. Intuitively, it measures how much tree-like a given graph is. Many problems which are NP-hard on general graphs become tractable on graphs of low treewidth. However treewidth quickly hits its limits once we try to apply it to directed graphs. Directed acyclic graphs (DAGs), on which many problems have simple efficient algorithms, can have arbitrarily high treewidth. Naturally one can ask whether there is a digraph width measure with all the nice properties of treewidth. In this talk we first quickly survey some of the known digraph width measures, and then try to answer the question whether there indeed is a good directed counterpart to treewidth.
10. 10. 2017
Doc. RNDr. Gabriel Semanišin, Ph.D., dean of Prírodovedecká fakulta UPJŠ Košice
Two challenges in networks communications
Abstract: Internet of things and an enormous recent expansion of social networks have brought a lot of new challenges in the areas of secure and efficient communications in networks.

In the talk we will discuss two aspect of such communication. We will provide some motivation and background of the treated problems and later we will concentrate on two specific challenges that are related to the creation of the network topologies and afterwards to the monitoring of the communication in the networks.

We will briefly demonstrated two graph theory methods based approaches to the above problems. One is related to the Load Balancing Problem and the second on to the k-Path Vertex Cover Problem that provides a generalization of the well-known Minimum Vertex Cover Problem.

Finally, we will briefly present goals and outcomes of our two recent international pilot projects in electronic communications at which we have been taking part.

17. 10. 2017
Dr. rer. nat. Achim Blumensath, FI MU
Algorithmic Model Theory
Abstract: Tools from logic are used in many areas of contemporary computer science including databases, verification, and algorithms and complexity. The field of algorithmic model theory developed in response to these applications. It studies the various logics used by applications with a focuses on two related aspects: (i) algorithmic questions, in particular algorithms for the model checking problem; and (ii) questions concerning the expressive power of these logics.

In this talk I will give an overview over the various aspects of algorithmic model theory and then concentrate on a few selected areas. A recurring topic will be the notion of a compatible operation. We will discuss several applications of such operations throughout algorithmic model theory. As a more detailed example I will present a classification theorem for a particular kind of operations called a transductions.

24. 10. 2017
Mouzhi Ge, Ph.D., FI MU
Overview and New Trends in Data Quality Research
Abstract: Numerous business initiatives have been delayed or even cancelled, citing poor-quality data as the main reason. Data quality has become a critical concern to the success of enterprises. Due to the importance of data quality, the research of data quality and its management strategies can already be traced back to the 90s. Thus, I will firstly provide an overview of the data quality research over the last 20 years and discuss the development of data quality research topics such as data quality assessment, data quality management, and data quality in contexts. I will further refine two state-of-the-art research trends: one is Big Data Quality, which will be explained in the context of smart city. The other is Relative Data Quality, which will be applied in the enriched geospatial applications. Also, the talk will include a short introduction to the data quality research community, the industrial applications, and how to apply data quality research in other research domains.
31. 10. 2017
prof. ing. Jana Košecká, Ph.D., George Mason University, Google Brain, Mountain View, CA, USA
Advances in Learning for Robots Perception
Abstract: Advancements in robotic navigation, mapping, object search and recognition rest to a large extent on robust, efficient and scalable semantic understanding of the surrounding environment. In recent years we have developed several approaches for capturing geometry and semantics of environment from video, RGB-D data, or just simply a single RGB image, focusing on indoors and outdoors environments elevant for robotics applications.

I will demonstrate our work on predicting locations of generic objects in videos acquired by a moving vehicle, for detailed semantic parsing using deep convolutional neural networks (CNNs), object detection and object pose recovery from single RGB image. The applicability of the presented techniques for autonomous driving, service robotics, mapping and augmented reality applications will be discussed.

7. 11. 2017
doc. RNDr. Jan Strejček, Ph.D., FI MU
Loop summatization and its applications
Abstrakt: We show a symbolic-execution-based algorithm computing the precise effect of a program cycle on program variables. For a program variable, the algorithm produces an expression representing the variable value after the number of cycle iterations specified by parameters of the expression. The algorithm is partial in the sense that it can fail to find such an expression for some program variables (for example, it fails in cases when the variable value depends on the order of paths in the cycle taken during iterations).

We present two applications of this loop summarization procedure. The first is a construction of a nontrivial necessary condition on program input to reach a given program location. The second application is a loop bound detection algorithm, which produces tighter loop bounds than other approaches.

14. 11. 2017
pplk. doc. Ing. Jan Mazal, Ph.D., NATO Modelling & Simulation Centre of Excellence, Rome
Military robotics and the Army of the Czech Republic
Abstract: The world of information technology and automation passed an unprecedented expansion in the last decade and it affects all areas of our lives today, including the military. As shown in many military operations, for example in Iraq, Afghanistan, Syria, and other countries, the results of research and development have played a decisive role, indicating that the high technological approaches in combat has proven itself as a key element of a victory in almost all conducted conflicts.

This factor is speeding up the development of selected military-technological areas, which slowly indicate a breakthrough in the ways of the combat operations management in "the depth and width", which the human has not experienced throughout its existence yet. With a high probability it can be said, that we are on the edge of another technological revolution, which certainly will have a serious impact on the global society. It will perhaps surpass all previous milestones of human progress and maybe even some visions from Sci-Fi literature.

The impacts of significant advances in the development on the field of artificial intelligence, robotics and industrial automation, can be very difficult to predict, even in the short term.

The aim of the presentation is to briefly inform students with a short history of armed conflicts and the impact of technology on command and control, on possible approaches that lately come, also, possible ways, how to implement robotic systems in military operations and the effect of artificial intelligence on the military operations management, how to increase the effectiveness of operational command and control, and briefly summarize research and development in the field of military robotics within the Army of the Czech Republic.

Short biography: LTC. Doc. Ing. Jan Mazal, Ph.D., graduate of the Faculty of Military Systems Management of the Military College of Ground Forces in Vyskov. In 2003 he graduated the Academic Course of Military Intelligence in Fort Huachuca, Arizona, USA. Since 2005 he is a doctor in the field of the theory of the Defence Management of the State and since 2013 he is associated professor in the problematic of military management and C4ISR systems. He is former deputy chief of the Department of Military Management and Tactics at the University of Defence in Brno, currently he works as Doctrine Education and Training Branch Chief at NATO Modelling & Simulation Centre of Excellence in ROME. His field of research and expertise is military intelligence and reconnaissance, C4ISR systems, and Modelling and Simulation as a military decision making support. He is the author or co-author of more than 70 professional papers/publications, he solved more than 10 scientific projects, he is the author of a number of robotic prototypes and application software. In his previous military practice, he also held command and staff functions and also he took part in the International Military Operations as EUFOR (Bosnia and Hercegovina) and ISAF (Afghanistan).

21. 11. 2017
RNDr. Martin Maška, Ph.D., FI MU
Segmentation and Tracking of Spiky Cells
Abstract: To monitor their surroundings and to promote their movement, motile cells protrude spikes. Although modern fluorescence microscopy allows us to observe spiky cells at unprecedented spatio-temporal resolutions, a fully 3D image analysis approach to automatically quantifying the behavior of spikes is still lacking. In this talk, we briefly explain crucial aspects of such a quantitative task and present our recent advances in the field, demonstrating how convolutional neural networks combined with realistic computer-generated image data help to increase segmentation and tracking performance beyond that of traditional image analysis approaches without machine learning.
28. 11. 2017
Univ.-Prof. Dr. Uwe Zdun, Faculty of Computer Science, University of Vienna
Software Architectures and Techniques for Compliance Monitoring and Enforcement
Abstract: IT compliance means in general complying to regulations that apply to an IT system. There are many other rules and constraints in a software system that have characteristics similar to IT compliance rules stemming from regulations, including security policies, business rules, QoS rules, deployment rules, and even architecture conformance rules. In this talk, we discuss software architectures and techniques for automated compliance monitoring and enforcement. Goals are to enhance the automation in compliance controls and make compliance controls easier to implement, reuse, and change - and thus less costly. We will discuss different categories of compliance checking techniques, like those that check at design time, runtime, or after runtime, and those that are primarily focussed on structures or behaviors. Further we discuss the need and techniques for involving domain experts more closely, as well as semi-automatic guidance for control definition, monitoring, and enforcement.
5. 12. 2017
doc. RNDr. Tomáš Brázdil, Ph.D., FI MU
Mastering games with deep learning
Abstract: The main theme of the talk is how much a deep learning agent could learn by playing games against itself. I will start with a historical overview of TD-Gammon, a famous algorithm for playing backgammon at expert level. In the main part I will concentrate on the very recent algorithms for playing Go above master level. I will give an overview of fundamental methods, namely deep reinforcement learning and Monte Carlo tree search. Finally, I will comment on new directions in the area, especially the Starcraft II challenge in which the current algorithms still fail to deliver convincing results.
19. 12. 2017
prof. Stefan Bruckner, Faculty of Mathematics and Natural Sciences, University of Bergen
Navigating the Space of Visualizations
Abstrakt: Considering the vast amounts of data involved in many scientific disciplines and industrial applications, it is essential to provide effective and efficient means for forming a mental model of the underlying phenomenon. The term "visualization" refers to the process of extracting meaningful information from data and constructing a visual representation of this information.

While the concept of using images to communicate complex phenomena of course predates the development of digital technology by millennia, over the past decades the field of visualization has firmly established itself as an important and constantly expanding discipline within computer science. Computer-based Visualization seeks to provide interactive graphical data representations, taking advantage of the extraordinary capability of the human brain to process visual information. Advanced visualization methods now play an an important role in the exploration, analysis, and presentation of data in many fields such as medicine, biology, geology, or engineering. This development, however, has also lead to the fact that there is now a vast number of often very specialized techniques to visualize different types of data tailored towards specific tasks. For non-experts it becomes non-trivial to choose appropriate methods that will provide the optimal answers to their questions.

In this talk, I will discuss previous and ongoing research on how we can explore and navigate the space of visualizations itself. By consider the interplay between data, visualization algorithms, their parameters, perception, and cognition as a complex phenomenon that deserves study in its own right, we are making progress in providing goal-oriented interfaces for visual analysis. For instance, we can make the modification of input parameters of visualization algorithms more intuitive by normalizing their perceived effects over the entire value range, and provide visual guidance about their influence. Furthermore, by incorporating additional knowledge into the visualization process, we can infer information about the goals of a user, and develop smarter systems that automatically suggest appropriate visualization techniques. This line of investigation leads us along the path towards a new type of visual data science, where automated data analysis approaches such as deep learning are tightly coupled with interactive visualization techniques to exploit their complementary advantages for knowledge discovery in data-driven science.

Short biography: Stefan Bruckner is professor in visualization at the Department of Informatics of the University of Bergen, Norway. Prior to his appointment in Bergen, he was an assistant professor at the TU Wien, Austria where he also received his habilitation (2012) and PhD (2008). His research focuses on interactive visualization and he has made important contributions to several areas such as illustrative methods, parameter space exploration, feature detection, and knowledge-based interfaces. Prof. Bruckner has successfully led several research projects including industry collaborations with partners such as AGFA HealthCare and GE Healthcare. His results were published in the premier venues for visualization research and have to date received 8 best paper awards and honorable mentions at international events. He won the Karl-Heinz-Höhne Award for Medical Visualization and received the prestigious Eurographics Young Researcher Award. As an active member of the international scientific community, Prof. Bruckner regularly serves on the program committees of the leading conferences in visualization and computer graphics. He was program co-chair of EuroVis, PacificVis, VCBM, and the Eurographics Medical Prize, and currently serves on the editorial board of Computers & Graphics. He is a member of the IEEE Computer Society, ACM SIGGRAPH, and Eurographics.