Research work in the field of Artificial Intelligence and Machine Learning
The research goal is to develop methods that allow targeted manipulation of complex content, such as speech, images or video. The context of the work is crowned with many spectacular successes of research on generative algorithms using deep neural networks. In the course of previous work, it was possible to formulate and successfully implement the concept of functional control over modifications of qualitative image and video attributes, summarized in articles [1, 2]. Current works are devoted, among other things, to the animation of images, made on the basis of single photos, and the formation of a style of synthetic expression.
Contributors: Krzysztof Ślot, Paweł Kapusta, Kacper Kubicki, Jacek Kucharski
The aim of the work is to define, based on neural networks, one-way and multi-layer computing architecture composed of elements performing fuzzy logic functions. The context is the worldwide intensive research on the so-called "explainable artificial intelligence" (Explainable AI, XAI), and the result of the work is the concept of L-neuron [3] - an adaptive logical operator, which is a processing element of the network.
Contributors: Jacek Kucharski, Piotr Łuczak, Przemysław Kucharski, Krzysztof Ślot
The aim of the work is to enable the effective implementation of artificial intelligence algorithms in devices operating autonomously, with limited computing power and limited network connectivity (so-called edge-computing). The result of the work is the formulation of the concept of computational architecture [4,5] allowing for the implementation of complex analyzes and intended for implementation in specialized, energy-saving VLSI systems. The developed algorithms are currently being physically implemented as part of cooperation with foreign partners in the European grants Horizon 2020 (MISEL project) and CHIST-ERA (AiR project).
Contributors: Piotr Łuczak, Krzysztof Ślot, Jacek Kucharski
The aim of the research is to develop dedicated procedures for processing and analyzing digital images, based on both classic machine learning algorithms and deep convolutional neural networks. Their purpose is to automate the tasks of image analysis in specific domains, such as computer-aided medical imaging systems [6,7], analysis of ophthalmic images or supporting the analysis of geological [8] and dendrological images [9]. Based on generative models, techniques are also being developed to synthesize training image data for deep neural network models dedicated to image segmentation.
Contributor: Anna Fabijańska
[1] K. Adamiak, P. Kapusta and K. Ślot, 2020 International Joint Conference on Neural Networks (IJCNN), 2020.1-7. DOI link
[2] K. Ślot, P. Kapusta and J. Kucharski, Neural Computing and Applications, 2021. 33:1079–1090. DOI link
[3] P. Łuczak, P. Kucharski, T. Jaworski, I. Perenc, K. Ślot and J. Kucharski, Sensors, 2021. 21: 6168. DOI link
[4] P. Łuczak, K. Ślot K. and J. Kucharski, IEEE International Joint Conference on Neural Networks IJCNN, 2022. 1-8. DOI link
[5] K. Ślot, P. Łuczak and S. Hausman, Bulletin of the Polish Academy of Sciences: Technical Sciences, 2022. 70: e143552. DOI link
[6] M. Czepita and A. Fabijańska, Computer Methods and Programs in Biomedicine, 2021. 208: 106240. DOI link
[7] A. Kucharski and A. Fabijańska, Biomedical Signal Processing and Control, 2021. 68C: 102805. DOI link
[8] A. Fabijańska, A. Feder and J. Rigde, Computers & Geosciences, 2020. 144: 104584. DOI link
[9] A. Fabijańska A. and M. Danek, Computers and Electronics in Agriculture, 2021. 181: 105941. DOI link