My Theses:
Bachelor, Master, and PhD
My bachelor's and master’s theses
I studied Medical Physics at the University of Novi Sad, Serbia. Medical physics is most often associated with medical imaging or nuclear medicine, but my studies also covered the measurement of biomedical signals, such as brain waves (electroencephalography - EEG), or heart rate signals (electrocardiography - ECG), etc. We shared core physics courses with theoretical and condensed matter physicists, with additional subjects in medical imaging, nuclear physics, dosimetry, and biomedical measurements.
During my bachelor’s, I got interested in recording the electrical activity of brain signals and their data analysis and interpretation. This interest led me to focus my bachelor's thesis on electroencephalography, specifically the P300 EEG component associated with cognitive processing and attention. I worked on data collection and analysis, exploring how external stimuli influence brain activity (see Figures 1 and 2).
Figure 1: Putting an EEG headset to one of the participants.
Figure 2: Averaged P300 component (red) and “regular” EEG measurements (green).
The thesis is written in Serbian, with abstracts in Serbian and English. You can find the thesis here and here.
After earning my bachelor's in Medical Physics, I shifted my focus to data analysis and the measurement of biomedical signals. For the thesis, I developed virtual instruments for analyzing electrophysiological signals using LabVIEW (see Figure 3).
My master's thesis is also written in Serbian, with abstracts in Serbian and English. You can read it here.
Figure 3: Part of the LabView code for processing electrophysiological measurements.
My PhD thesis
(text adapted from the Introduction of my PhD thesis)
I started my Ph.D. studies at the Institute of Cognitive Science in Osnabrück, Germany (Figure 4) after finishing my bachelor’s and master’s studies. Since I had previously worked with data analysis of EEG data, my initial interest was medical data in general and EEG data in particular. Following this direction, I wrote the first paper, “Predicting epileptic seizures using nonnegative matrix factorization.” During this project, we collaborated with Dr. Levin Kuhlmann from Monash University in Melbourne, Australia.
Figure 4: Me in front of the old building of the Institute of Cognitive Science, January 2015.
Following my interests in the interpretability of machine learning models and time-series analysis, I wrote my second paper, “A Bayesian Monte Carlo approach for predicting the spread of infectious diseases.” The paper is a result of the collaboration with Dr. Alexander Ullrich and Dr. Stephane Ghozzi from the Robert Koch Institute in Berlin. As a part of the project, I spent two months at the Signale Group of the Robert Koch Institute, working on implementing the proposed model. During this time, I learned a lot about Bayesian methods and their potential for developing interpretable machine learning models. Further, I studied how interpretable models help communicate scientific results to a broader audience. As a part of the project, I presented a poster at the EU Data Viz conference in 2019 titled “Visualizing the spread of infectious diseases using public health data” (Figure 5). The poster presented our developed methods and how they could serve citizens through data visualization and communicate epidemiological data to the public.
Figure 5: Presenting my poster at the EU Data Viz conference, November 2019.
Continuing my interest in the Bayesian approach, I wrote the third paper, “Bayesian hierarchical models can infer interpretable predictions of leaf area index from heterogeneous datasets.” It explores the possibilities of Bayesian hierarchical models for developing interpretable machine learning models for application in remote sensing and, more broadly, environmental sciences. In this project, we collaborated with Dr. Bastian Siegmann from Jülich Research Centre and Dr. Thomas Jarmer from the Institute of Computer Science in Osnabrück. Here, I used Bayesian hierarchical models to deal with challenges similar to those I encountered in previous projects: making predictions from limited and heterogeneous datasets.
You can find my PhD thesis titled “Interpretable machine learning for real-world applications“ here.