Dr. Joanna Komorniczak from our Faculty received an award in the 15th Polish Artificial Intelligence Association Competition for the Best Polish Doctoral Thesis in the Field of AI. The jury appreciated her dissertation on using meta-features in processing time-varying data streams and detecting concept drifts. Congratulations!

Our researcher from the Department of Computer Systems and Networks, Dr. Joanna Komorniczak, was distinguished for her doctoral thesis titled “Non-stationary data stream processing with meta-feature analysis” supervised by Prof. Paweł Ksieniewicz.

Her work deals with processing data streams, with particular emphasis on time-variable data, i.e. on data subject to the so-called concept drift.

“In my doctoral dissertation, I proposed methods for detecting concept drifts and solutions for classifying non-stationary data streams. They include methods designed for recurring concepts, typical of cyclical phenomena, where data distributions depend, for example, on the time of day or the season of the year. The study also highlights the limitations associated with assessing the performance quality of such methods in synthetic experimental environments,” explains Dr. Komorniczak.

As she explains, stream processing is used in machine learning systems that require an immediate response, continuous quality monitoring, and updating of the possessed knowledge. In such conditions, the nature of the analysed problem can evolve. Changes may include, among others, the proportions of individual classes or decision boundaries used by the model, which may lead to a deterioration in recognition quality.

Detection of such changes typically requires direct monitoring of the system's operation and the use of expensive and hard-to-access labels. Solutions developed by the researcher are based on meta-characteristics, which are measures describing data distributions at a general level. They show the drift of the concept and react to it appropriately even before a noticeable drop in the quality of classification.

“It is important for me that my dissertation was distinguished in the competition, as it summarizes several years of my research, during which I learned not only how to design machine learning methods and solutions but above all how to verify their operation. I am pleased that the PSSI committee has recognized the issue of data stream variability as significant and has appreciated solutions based on meta-features analysis,” says Dr. Komorniczak.

The main prize in the competition was awarded to Dr. Łukasz Maziarka for his dissertation “Adapting Deep Learning Architectures for Drug Discovery”, defended at the Jagiellonian University. His supervisor was Dr. Jacek Tabor, with Dr. Stanisław Jastrzębski as the assistant supervisor.

Alongside Dr. Joanna Komorniczak, an honorary award also went to Dr. Kamil Faber for the work “Towards Lifelong Anomaly Detection in Challenging Scenarios and its Application in Cybersecurity”, defended at the Stanisław Staszic AGH University of Science and Technology in Kraków. His supervisor was Prof. Bartłomiej Śnieżyński.

The winners of the 15th edition of the competition will be invited to present the results of their research during the Symposium of the Polish Artificial Intelligence Association, which will take place on 24-25 September, 2026, at Gdańsk University of Technology.