Welcome to my homepage!
Home — Research — Teaching — Academic services — Projects — Education — Work Experience
My broad research interests include everything related to the application and theory of Machine Learning. Throughout my research career, I gained experience in the following topics: recommender systems, genetic optimization, application of Machine Learning techniques in security and education, building robust systems using conformal learning, causality in Machine Learning. Below I describe what I have worked on so far. You can find out more from my publications by checking my Google Scholar profile.
During my PhD studies, I worked on the explainability of matrix factorization-based recommender systems. Matrix Factorization is proven to be an effective approach for generating personalized recommendations. The major drawback of this method is the lack of explainability. We proposed to align the latent factors of matrix factorization-based recommender systems with a set of users/items. The interests/characteristics of the latter can be considered as the building blocks of the factorization model. This approach not only allows interpreting the generated recommendations but also results in an elegant solution to the cold-start problem. Read Part I of my thesis and the relevant publications to find out more.
In Part II of my thesis, we propose a method to identify so-called trigger factors. These are the factors that can influence the values of other factors in the system. Such factors can be considered as the building blocks of the system and can be useful in both prediction and prescription tasks. Lately, I am working on aligning the proposed approach with methods from causal machine learning.
Within several European projects, I studied the security aspects of privacy-enhancing communication systems. Recent investigations show that it is possible to deanonymize users of such systems through the analysis of metadata. We identified threats in a newly proposed modification of the TOR browser and presented them in this publication.
Within a project BacAnalytics we collaborated with the (Academy of Nancy-Metz)[http://www.ac-nancy-metz.fr/] of the French region Grand Est to propose a Machine Learning-based solution to organize the final school examination. This project was awarded the French national prize Impulsion 2018 in nomination “Innovation”. The details can be found in this publication or the relevant video presentation.
In our recent work, we proposed an alternative multiobjective sorting procedure to overcome some problems of Pareto dominance-based genetic optimization algorithms. The relevant work was accepted for presentation at GECCO-2021.
My recent activities also include research in conformal learning. More precisely, we have been studying the impact of different model-agnostic nonconformity functions on the efficiency of conformal predictors. We also proposed a novel strategy for the aggregation of conformal predictors that allows improving their characteristics. This work is currently under revision.
Currently, I am also exploring causal machine learning. I authored a relevant course at the University of Luxembourg and several workshop and tutorial sessions, see my teaching experience. My research activities in this direction include studying the impact of the noise level in the Additive Noise identification approach.