Robustness Studies for GNN-tracking at Belle II

This thesis examines the robustness of the CDC AI Tracking (CAT) algorithm, developed for track reconstruction at the Belle II experiment using Graph Neural Networks (GNNs). As the luminosity of the SuperKEKB collider increases, traditional track reconstruction methods face challenges such as reduced wire efficiency and malfunctioning readout boards. The CAT algorithm is designed to adapt to these issues, offering improvements in tracking performance, particularly in detecting displaced tracks under suboptimal conditions. This work evaluates the effectiveness of the GNN-based approach in overcoming these hardware-related challenges and compares it to the traditional Legendre-based method. The results highlight the advantages of the GNN approach in maintaining accuracy and robustness, even in the presence of hardware failures.