News

We invite you to the defense of the dissertation of Ing. Ondřej Olšák from the Department of Computer Systems, FIT BUT, which will take place on Thursday, February 5, 2026, at 1:00 p.m. in meeting room C209. The supervisor of the dissertation entitled "Acceleration of wave propagation simulation calculations using pruned Fourier transform" is Prof. Jiří Jaroš.

Ondřej Olšák's research is closely linked to possible practical applications, particularly in the highly monitored field of medicine. In his work, he focuses on optimizing wave propagation simulations, which are key to planning neuromodulation procedures and tissue ablation using ultrasound waves. The SC@FIT research group led by Prof. Jiří Jaroš, of which Ondřej Olšák is a member, focuses not only on the development and optimization of wave propagation simulators at FIT. The group is involved in the development of the open-source toolkit k-Wave, which is one of the most highly regarded tools in the field of wave propagation simulation. k-Wave uses a k-space pseudo-spectral method based on Fourier transformation to calculate wave propagation simulations, which allows for high computational accuracy. However, the Fourier transform calculation accounts for approximately 60% of the total simulation computation time and presents a challenge for optimization.

Simulating wave propagation in high-resolution domains is a computationally demanding task—a single simulation can take from seconds to hours or days, depending on its parameters. When planning medical procedures, it is necessary to perform dozens of such simulations. This is where the key challenge behind Ondřej Olšák's research comes in: to find a new optimization approach to speed up wave propagation simulations using spectrum pruning techniques. When asked for a layman's explanation of the research goal, Ondřej Olšák begins broadly: "In order to simulate the propagation of ultrasound waves in the human body, we must convert the real structure of tissues into digital form—into a regular grid of points. Each point carries information about the type of tissue at that location. The finer the grid we use, the more accurately we can capture the actual shapes and transitions between different types of tissue, and the more accurate the result of the simulation itself will be. However, as the fineness of the grid increases, so does the computational complexity. The key to optimization was understanding how the coefficients in the spectral (frequency) domain of the propagated wave change and what influences their position. After conversion using a fast Fourier transform, the wave is mathematically expressed as a set of spectral coefficients. When we increase the resolution of the domain, the spectral region will contain more of these coefficients. However, for high-resolution domains, we do not need to calculate all the coefficients that occur in the spectral domain to achieve an acceptable result. When calculating the Fourier transform, we can therefore neglect certain parts of the spectrum and calculate only those coefficients that are truly important for the result. This technique, called spectrum pruning, allows for a significant acceleration of the simulation while maintaining acceptable accuracy.

This is the core of the optimization that Ondřej Olšák designed and experimentally verified for the k-Wave tool. The result of his work is a modification of the existing k-Wave implementation for both 2D and 3D simulations. Experiments performed on anatomical models of the human head and liver with different sizes of computational domains – up to a resolution of 9216×12288 points for 2D and 567×672×448 points for 3D – showed acceleration of up to 1.9× for two-dimensional and 1.7× for three-dimensional simulations. However, the spectrum pruning technique comes with a trade-off: neglecting part of the spectral coefficients introduces a certain error into the calculation. The key is to find the right balance between the range of the neglected part of the spectrum and the acceptable accuracy of the simulation results. Determining the acceptable error limit is not a trivial task and provides an incentive for further research in this area.

According to Olšák, the above-described optimization of calculations can be used, for example, in situations where the appropriate position of the ultrasound transmitter is sought before the actual procedure, so that the ultrasound waves are focused as accurately as possible on the desired part of the brain or other tissue and no damage is caused to the surrounding tissue. In this process, it is possible to use an optimized version of the simulation to quickly explore different configurations, and once the appropriate position of the transmitter has been found, to perform a control calculation using a non-optimized simulation to ensure maximum accuracy of the result. This procedure thus significantly reduces the time and costs required for treatment planning. It should be added that the results of Olšák's research can be applied not only in the medical field, but also in other areas requiring wave propagation simulations in sparse heterogeneous media.

Ondřej Olšák takes a pragmatic view of his doctorate and its imminent completion: "I like the fact that someone will now read my work as a whole and give me further feedback on my research. I was already involved in the research during my master's studies. I enjoyed searching for solutions and testing hypotheses the whole time." He himself admits that it was not always an easy journey: "At the beginning of my dissertation research, I reached a dead end and felt like I was at a loss. Then one day, a video I happened to see gave me an idea for a different method. It was a wonderful, indescribable feeling." He would like to continue his research, but his next steps will be in industry, where he sees interesting challenges. "I would like to thank my family for their support throughout my studies. And then, in addition to my supervisor, Professor Jaroš, I would like to thank all my colleagues from the research group for their long-term help and support," Ondřej Olšák recalls with gratitude at the end of our interview. We will keep our fingers crossed for him, and not just for his public defense of his dissertation. We wish him the joy of discovering new paths in his professional practice as well.

We invite you to the defense of the dissertation of Ing. Martin Hurta from the Department of Computer Systems, FIT VUT, which will take place on Wednesday, January 21, 2026, at 10:00 a.m. in meeting room G108. The supervisor of the dissertation entitled "Advanced Cartesian Genetic Programming for Biomedical Applications" is Prof. Lukáš Sekanina.

Martin Hurta has long been involved in the field of Cartesian genetic programming (CGP), an artificial intelligence technique inspired by natural evolution. Genetic programming was historically developed for the purpose of designing electronic circuits, but today it is often used in connection with the design of FPGA-type embedded circuits. This method belongs to a category of artificial intelligence algorithms called evolutionary algorithms. There are a number of subcategories of nature-inspired algorithms, evolutionary algorithms being one of them (others include artificial neural networks). The essence of Cartesian genetic programming is the automated design of algorithms, where, in layman's terms, we have several candidate random solutions, run and evaluate them all, select the best one, and create new solutions through random changes. By repeating these operations, the solution is gradually improved, following the model of evolution, until the result is a working program. We can then imagine the solution of Cartesian genetic programming as a graph of operations working with inputs and constants. This graph takes the form of a 2D grid of nodes, which is internally represented by a string of values. The above-mentioned changes in the connections between nodes and the functions they perform take place in the grid. Following the example of genetics, experts refer to these as solution mutations.

The scalability of evaluation, i.e., the evaluation of all candidate solutions in hundreds of thousands of iterations, is a time- and energy-consuming challenge that Martin Hurta is trying to solve in his research. Although Cartesian genetic programming is capable of proposing very interesting solutions in tasks such as classification, symbolic regression, and circuit design, its real-world application is still limited. This is particularly evident when compared to machine learning techniques such as artificial neural networks. The reason for this, apart from the computational complexity itself, is also the unfamiliarity with these procedures outside the narrow community of researchers.

In his dissertation, Martin Hurta explores possible modifications and adaptations of Cartesian genetic programming that would help improve the properties of both the algorithm itself (design time, computational requirements) and the proposed solutions (classification accuracy, ratio between accuracy and hardware requirements, explainability). He then experimentally verifies the proposed methods on problems in the field of biomedical informatics and bioinformatics, which can benefit from the explainability of the proposed solutions or the possibility of their implementation in energy-efficient wearable devices (e.g., medical monitoring systems) – something that conventional machine learning methods have difficulty with. Hurta specifically mentions his collaboration with Prof. Stephen Smith from the University of York: This project uses devices that monitor the movements of patients with Parkinson's disease, who often suffer from dyskinesia (in very layman's terms: involuntary movements), using an accelerometer and gyroscope. The goal is to automatically recognize the manifestations of dyskinesia in patients' movements and enable doctors to use this information to correctly target medication (the drug Levodopa) and its dosage. Similarly, Hurta participated in the application of Cartesian genetic programming algorithms to recognize alcohol abuse and depressive disorders from EEG, and to calculate polygenic risk scores and predict plant growth based on genetic information (here it is worth mentioning his collaboration on calculations with the Institute of Biomedical Engineering at FEKT, Wolfram Weckwert from the University of Vienna, and Dirk Walther from the Max Planck Institute of Molecular Plant Physiology). It should be added that polygenic risk scores generally allow the risk of predisposition to a negative phenomenon to be determined on the basis of multiple genes; in humans, this most often involves determining the likelihood of a specific disease. When asked to identify the field of application he is most intensively involved in and favors, Hurta's choice is clear: "Parkinson's disease is the one I've been working on the longest and is also very topical. My colleagues and I are continuing to work actively on this application, and I believe that the proposed solutions could become part of diagnostic tools in the future and thus really help those affected. In addition, it is possible to consider a number of other diseases that affect human movement in the future."

There are many potential practical applications, and according to Martin Hurta, there is a reason for this. "There are so many methods of artificial intelligence and machine learning, and practically each one has its own possible applications, a field in which it can shine. The advantage of Cartesian genetic programming is that its solution can be written as a mathematical equation. And there are areas where this is critical—often in medicine, the space industry, military technology, and so on. Thanks to the mathematical equation, experts in these fields can say, 'It may be artificial intelligence, but it has designed this specific solution, this equation, and that is what will go into our device. It won't just be some kind of black box,'" says Hurta, describing the key advantage of genetic programming. He adds: "It would be great if our classifiers were to find their way into other medical 'boxes'."

At FIT, Martin Hurta is part of the research group EvoAI Hardware, led by his supervisor, Prof. Lukáš Sekanina. "I see a doctorate as a natural step. I want to continue at the faculty. I am part of a new large project led by Dr. Vojtěch Mrázek, within which I can continue working on similar topics. I'm glad to have a 'stamp' on my progress; people are always happy to have their position secured," says Martin Hurta, commenting on his current situation and possible future with a parallel to computer games.

We wish Martin Hurta a successful defense and much success in his future research.

Would you like to improve your skills in working with AI tools? The European project VASSAL is organizing two workshops at our faculty focused on the practical and effective use of AI tools a) in administrative and IT work, b) in scientific research. The program is divided into two full-day workshops on January 15 and 16, alternating between morning (9:00 a.m.–12:00 p.m.) and afternoon (1:00 p.m.–3:00 p.m.) blocks. Both training sessions will be led by Prof. Daniel Mertens, biochemist, lecturer, and head of research groups at the German Cancer Research Center (Das Deutsche Krebsforschungszentrum) and the University of Ulm.

The first workshop focuses on the use of artificial intelligence (primarily in the sense of large language models) in administrative and IT work. Participants can expect a specific agenda of artificial intelligence applications for meeting and correspondence preparation, data and project management, and simplified reporting. The second workshop focuses on the use of LLMs in coding (vibe coding) and for scientific creativity/brainstorming and communication. The event is aimed at researchers, Ph.D. students, technical and economic staff, and generally anyone interested in the everyday use of AI tools.

Prof. Mertes' courses are based on specific practical experience and joint problem solving. The workshops are free, but registration is required.

For more information, visit the VASSAL project website.