PRELUDE 18 grant for a PhD student from the Faculty of Electrical and Control Engineering | Gdańsk University of Technology

In the latest competition of the National Science Center PRELUDIUM 18, intended for young scientists without a doctoral degree, funding was obtained for the research project Detection and reduction of the impact of data bias using explanatory artificial intelligence, proposed by MSc Eng. Agnieszka Mikołajczyk from the Faculty of Electrical and Control Engineering.

The three-year scientific project will be a continuation of the current research of a PhD student from the Department of Electrical Engineering, Control Systems and Informatics, which since its engineering work deals with the use of artificial intelligence methods, in particular neural networks. Project research will be devoted to data bias in collections on which neural network learning is based.

Artificial neural networks are taught on large collections of specially prepared data, the quality of which has a huge impact on the effectiveness of inference models using them. A common problem is bias, the selection of which is not sufficiently representative. To estimate the possible errors of inference models related to data bias, an in-depth analysis of the models is necessary, for which techniques from the so-called Explainable Artificial Intelligence (XAI).

As part of the project, it is planned to develop methods to detect data bias and, more importantly, eliminate their impact on the learning process of neural networks and the operation of inference models. The project envisages the integration of XAI methods with existing and new artificial intelligence systems, in particular with deep neural networks. The developed methods are to be tested in applications in the field of computer image analysis.

The supervisor of the doctoral thesis is DSc Eng. Kazimierz Duzinkiewicz, university professor and the supporting supervisor is DSc Eng. Michał Grochowski.