Input Feature Optimization for Track Reconstruction using Graph Neural Networks at Belle II

This thesis is describing the optimization of the input features of the CAT finder for the tracking at Belle II. The CAT finder, developed by the Belle II tracking group, is using GNNs to track charged particles in the CDC of the Belle II detector. The goal of the thesis is to optimize this Neural Network by applying different methods of input feature optimization, for a shorter training time and a higher performance. After the investigation of the input features, different models are trained and evaluated with different methods of input feature optimization. The results are then compared and in the end a best configuration model is trained and compared to the standard CAT model and the Baseline.