Daniel Banciu

Investigated the predictive capabilities of FourCastNet, a deep learning weather prediction model, focusing on its utilisation of the Madden-Julian Oscillation (MJO) for enhancing weather forecast accuracy. Analysed predictive performance metrics such as root-meansquare-error and anomaly correlation coefficient.

Developed a critical feature for the vCharM system by Vector Informatik GmbH to enhance load management in electric vehicle charging networks. This feature involved gathering and analyzing vehicle charging data to refine load management algorithms, thereby optimizing power grid operations in residential and company settings.

Development of an application for the Bundesgartenschau 2019 in Heilbronn as part of the big project in the 5th semester of the bachelor's degree.

Challenges involved imitation learning and reinforcement learning using the car gym environment.

Analysis of an unknown data set in a project in teams of two. We analysed every speech in the Bundestag and tried to answer questions like "How has the language developed?"

As part of the course "Praktikum Bildverarbeitung, Maschinelles Lernen und Computer Vision" we show how we developed a pytorch implementation of the paper ’Toward fast and accurate human pose estimation via soft-gated skip connections’.

Evaluation the effectiveness of attribution methods for tabular data. Comparing the performance of SHAP and LIME attribution methods. Using a transformer-based language model, known as GReaT.