Efficiency Improvements Using Machine Learning in Event Generators for the LHC

In particle physics, event generators play an important role in connecting theoretical predictions with experimental results. Especially for experimental issues one often wants to generate events which follow a probability distribution based on the underlying physics. The generation of such events is currently done by a 'hit-or-miss' method, which is particularly for processes with many final-state particles very expensive in terms of CPU time. In this thesis a modified variant of this algorithm is introduced, which combines the 'hit-or-miss'-method with machine-learning techniques in order to improve the performance. Several possible applications of this new method are presented and compared with currently available methods of event generation in SHERPA for different processes.