The Standard Model of particle physics has proven to be remarkably successful in describing the fundamental constituents of matter and their interactions. Nevertheless, several open questions remain unanswered, motivating the development of future high-precision particle physics experiments and collider facilities. Among the proposed next-generation accelerators, the CERN Future Circular Collider in its electron–positron mode, FCC-ee, offers unprecedented opportunities for precision measurements of electroweak observables and detailed studies of the properties of the Z boson, the Higgs boson, and heavy quarks.A key challenge in these studies is the identification and discrimination of quark–antiquark final states. Accurate quark flavor tagging is essential for improving the precision of Standard Model measurements and enhancing the sensitivity to possible signals of physics beyond the Standard Model. While machine learning techniques have already demonstrated significant potential in jet tagging applications at hadron colliders, their application to future electron–positron collider environments remains comparatively unexplored. The availability of detailed detector simulations and large samples of simulated collision events for future collider concepts now provides the opportunity to investigate novel approaches for quark–antiquark discrimination in a way that has not previously been studied.The goal of this thesis is therefore to improve quark–antiquark tagging for electron–positron collisions using modern deep learning techniques. In particular, this work focuses on simulated e+e− → Z → q¯q events corresponding to the FCC-ee detector environment. Different neural network architectures are investigated and compared in several classification tasks, with special emphasis on graph- and transformer-based approaches such as ParticleNet and Particle Transformer. By exploiting low-level particle information and advanced machine learning methods, this thesis aims to enhance the separation power between different quark flavours and improve the overall tagging performance.Improved quark flavour tagging would contribute directly to the precision physics program of future electron–positron colliders. Enhanced discrimination capabilities couldreduce systematic uncertainties in Standard Model measurements while simultaneously increasing sensitivity to rare processes and potential deviations from Standard Model predictions. Consequently, advances in machine learning–based tagging techniques may play an important role in enabling future discoveries and deepening our understanding of fundamental particle physics.
René Kootz (Wed,) studied this question.