000002347 001__ 2347
000002347 005__ 20210501181006.0
000002347 037__ $$aBELLE2-MTHESIS-2021-005
000002347 041__ $$aeng
000002347 100__ $$aCedric Ly
000002347 245__ $$aShower shape conversion between electrons and photons in Belle II using Cycle Generative Adversarial Network
000002347 260__ $$aHamburg$$bDESY$$c2021
000002347 300__ $$a97
000002347 500__ $$aPresented on 18 03 2021
000002347 502__ $$aMSc$$bHamburg, University of hamburg$$c2021
000002347 520__ $$aSimulation is essential to high-energy physics. The application of simulation ranges from development of software to the analysis of physical processes. For example the simulation of physical processes helps to gauge the expected outcome for specific rare decays of B mesons. The simulation in Belle II can be divided into two parts. The first part simulates the physics of the particle with given property like four momentum. The second part simulates the interaction of the simulated particle with the Belle II detectors. The quality of simulation is determined firstly by our understanding of physics and secondly by the consensus with measurements. Therefore, every discrepancy between Monte Carlo simulation (MC) and data could either be caused by a mistake in the measurements or by a lack of understanding of the fundamental processes and structure of the world. In many processes of interest, like ALP-strahlung or the B → Xq γ channel, the study is dependent on the quality of the simulation, especially for Belle II as a high precision experiment. Therefore, it is appropriate to look for new ways to avoid and even understand discrepancies between data and simulation. In this thesis a novel approach for simulating electromagnetic shower shapes in the Electromagnetic Calorimeter (ECL) of Belle II has been developed. A Cycle Generative Adversarial Network (CycleGAN) is a deep neural network architecture, which one could use to convert an element of one class into an element of another class. The overall goal would be to convert a real detected electron shower shape into a shower shape of a photon. The decisive property of a CycleGAN is the correlation between input and output. Meaning the generated photons will have the profile of a photon, but retain many features of the original input electron. In the ideal case, the conversion will retain the particle energy, the distinct background features, and only change (the shower shapes) features by the minimal amount needed, in order to recognize it as an element of the targeted class. An evaluation of the new approach will be presented. This approach may be able to produce calibration samples based on real detected particles and is not only limited by electrons and photons. This approach can be used in theory for any two classes and could enable in many different potential applications beyond the Belle II experiment.
000002347 700__ $$aDr. Torben Ferber$$edir.
000002347 8560_ $$fcedric.ly@belle2.org
000002347 8564_ $$uhttps://docs.belle2.org/record/2347/files/BELLE2-MTHESIS-2021-005.pdf
000002347 980__ $$aTHESIS