Presentation of the new rules for evaluation of academics at FI

High-Speed and Flexible Packet Processing on FPGA-based SmartNICs

The continuous growth of network traffic and the deployment of 100 Gb, 400 Gb and the evolution towards 800 Gb and terabit-class Ethernet links bring significant challenges for network monitoring, security and data centre applications. Traditional CPUs cannot sustain wire-speed packet processing at nanosecond timescales. This talk will focus on hardware acceleration using FPGA-based SmartNICs, which combine high throughput with flexibility. We will discuss time-critical operations such as packet parsing, classification, pattern matching and stateful connection tracking, and present hardware architectures we have designed to address them. Special attention will be given to the balance between performance and flexibility, including the role of P4 programmability compared to the configurability of specialised application profiles. Practical use cases such as DDoS mitigation, security monitoring and AI training will illustrate these approaches, with a concluding outlook on future terabit-scale networks.

The Lab Fest, held as part of the Informatics Colloquium, aims to introduce the academic community to the activities of research groups at the Faculty, specifically their staffing, areas of interest, involvement in grants, ongoing collaborations at the University, in the Czech Republic or internationally; future directions and current research or development results.

HCI: Human Computer Interaction Laboratory

AVELAB: A newly established research group at FI

Insights into Entity Component Systems

Entity Component Systems (ECS) are powerful data structures, often used in video games. In this talk I will present how this composition pattern works, what an ECS is, which parts it is made of, and typical operations it offers to its users. I will also explore architectural choices and optimization strategies tailored to different use cases.

In particular, the talk will cover the following topics:

What Entity Component Systems are;

How LLMs can be used to create them;

Different architectural choices and their performance impact;

How to optimize and parallelize an ECS;

And finally VECS: The open-source Vienna Entity Component System (+ a tool to visualize its content).

Unpacking the Degree: Hard Truths About Computer Science Education

With nearly 30 years in the trenches as a professional programmer-and half that time spent hiring and mentoring new graduates-I've developed some blunt conclusions about a Computer Science university degree. Some skills are taught well, others are only superficially addressed, and a few essential abilities are barely mentioned-or simply can't be taught in a classroom at all. If you believe a diploma alone prepares you for life in real-world software development, prepare for a challenging perspective.

This lecture will expose where academic Computer Science shines, and where it falls strikingly short-especially when it comes to preparing graduates for the sprawling complexity of modern enterprise codebases. I'll argue that no coursework or toy project can substitute for industry-honed skills like Test-Driven Development, and I'll point out that developing soft skills such as effective teamwork and clear communication is just as important as technical expertise. We'll also explore practical approaches for bridging the chasm between university and the realities of professional software teams. Whether you arrive ready to debate or simply eager to learn, bring your strong opinions or your curiosity-let's re-examine what Computer Science education truly needs to deliver.

Information Security in Digital Business and Intelligent Systems

Nowadays, digitalization and intelligent infrastructure change human activities and industrial systems. Disruptive technologies, such as cloud computing, blockchain, AI/ML systems, and others, have become applicable in various domains, including self-driving vehicles, e-health, innovative city applications, and industrial automated systems. The intensive use of these technologies also generates and manages a lot of data and information, which should be used for the intended purposes, made available when needed, and integral to making correct decisions. It means that security should be treated as the first-level citizen in the digitalised processes and intelligent infrastructures. This talk will consider the assets and values of digitalised systems and intelligent infrastructures. We will discuss their security weaknesses, harms, risks and countermeasures to mitigate them.

CoFI break

Sampling-based Motion Planning in Robotics: From Theory to Practice

Motion planning aims to compute feasible motions for robots, machines, drones, or even virtual characters in games. This talk introduces the fascinating family of stochastic motion planners - sampling-based planners - which search high-dimensional spaces and can solve problems across a wide range of applications, especially for systems with many degrees of freedom like robotic manipulators. Yet their performance can drop in surprising ways, such as when navigating tricky "narrow passages." We will show how vague or approximate knowledge about possible solutions and clever tricks can significantly boost their efficiency.

Symbolic Analysis of Probabilistic Loops without Sampling

Probabilistic loops are control flow constructs whose execution is influenced by random variable assignments and probability-based conditions. A key challenge in this domain is to automatically measure the uncertainty of such loops, especially when the number of iterations may be unbounded. While this task is notoriously undecidable in the general case, we identify a set of practical restrictions that make automated reasoning feasible without requiring manual guidance. Our method uses symbolic computation, combining algebraic techniques with known properties of probability distributions, to compute closed-form solutions for higher-order statistical moments of program variables across loop iterations. This presentation will showcase our approach in action through illustrative case studies.

Shielding for Safe and Fair Sequential Decision Making

In this talk, we will explore methods to enforce formally specified safety and fairness properties at runtime, an approach called shielded reinforcement learning. Shields rely on a model of the environment's behavior to analyze the safety of actions and prevent a learning agent from executing those that could potentially violate a formal safety specification. We will discuss how shields can be computed for environments that exhibit both probabilistic and adversarial behavior. Furthermore, we will present recent model learning approaches capable of deriving compact probabilistic models for high-dimensional environments, which can then be used to compute shields. Finally, we will show how similar techniques can be applied to assure group fairness properties, such as demographic parity, at runtime.

Vojtěch Suchánek: Lying to Hardware: Reverse-Engineering ECC on Smartcards

Elliptic Curve Cryptography (ECC) provides strong security with small keys, making it ideal for constrained devices like smartcards, though such devices remain vulnerable to side-channel attacks. Side-channel attacks often assume a white-box attacker who has detailed knowledge of the implementation. In contrast, real-world implementations of ECC are usually a black box. In this talk, we will close this gap for the attacker by designing a catalog of methods that can reverse-engineer the implementation details needed by the attacker. While motivated by cryptographic attacks, our focus will be on the mathematical tricks we play on hardware in order to extract the hidden information.

Ondřej Sotolář: Computational Social Science

The Interdisciplinary Research Team on Internet and Society (FSS & FI MU, irtis.muni.cz) studies phenomena at the intersection of social sciences and informatics. This talk introduces the field of computational social science and outlines how modern Natural Language Processing (NLP) methods open new possibilities for social inquiry. Representation-learning techniques make it possible to test social theories on large-scale textual data, providing computational evidence about communication patterns and social behavior. At the same time, insights from social science can enrich NLP models by grounding them in theories of language use, context, and social structure. We will illustrate this reciprocal relationship between NLP and social science through three recent studies from our group, highlighting both methodological advances and substantive findings.

Algorithms for drawing graphs with no or few crossings in the plane, on surfaces, and beyond

Planarity testing, the problem of deciding whether an input graph can be drawn without crossings in the plane, is known to be solvable in linear time since the 1970s. A distinctive feature of this problem is that it lies at the crossroad between topology and algorithms. In this talk, we will survey some natural and recent generalizations: Since some graphs are nonplanar, can we draw them without crossings in more general spaces, such as surfaces, or even more general spaces, obtained by gluing vertices, edges, and triangles together? If a graph cannot be drawn without crossings on a given space, how to draw it with as few crossings as possible? How can we untangle a graph drawn with crossings on a surface, if it is possible?

Lab Fest, held as part of the Informatics Colloquium, aims to introduce the academic community to the activities of research groups at the Faculty, specifically their staffing, areas of interest, involvement in grants, ongoing collaborations at the University, in the Czech Republic or internationally; future directions and current research or development results.

DISA: Laboratory of Data Intensive Systems and Applications

Sitola