• normalizing-flows
• higgs
• particle-physics
• cms
• lhc
• simulation
• deep-learning

The analysis of the Higgs boson decay to a bottom quark-antiquark pair heavily relies on the availability of a large number of simulated events to perform an accurate estimate of the background contribution. This represents a limiting factor for the analysis, which translates to a systematic uncertainty associated with the limited number of simulated events. However, the generation of more simulated events is limited by both the computing resources available to the experiment and the computational complexity of the current simulation algorithms. A different approach, based on Machine Learning techniques, is presented in this thesis. Starting from the generator-level information, the Flash Simulation approach is capable of simulating only high-level analysis observables by using Deep Generative Models. The model employed for the simulation belongs to the Normalizing Flow class of algorithms. A first prototype, containing only muons and jets, was already developed by the CMS Pisa group. In this work, proper and fake electrons associated with jets are added to the existing framework. Furthermore, this thesis demonstrates the effectiveness of Flash Simulation in reducing statistical uncertainties associated with the size of simulated samples.