Speaker
Description
After the discovery of the Higgs boson in the LHC experiments, measurements of its properties began. The measured mass of the Higgs boson is 125 GeV. At this mass, the probability of its decay into bb is about 58%, which means that more than half of the produced Higgs bosons decay into a pair of bb quarks. Therefore, this channel is very important for studying the properties of the Higgs boson. The most suitable production channel of the Higgs boson for decay to bb is the associated production with a vector boson. It was in this production channel that the decay of the Higgs boson into a pair of b quarks was first observed at the LHC in 2018. The Higgs boson is easier to study in the ZH associated production channel, where Z decays into a pair of charged leptons (electrons or muons), since the final state particles can be fully reconstructed in the detector. During the Run-1 of the LHC, physics analysis was carried out mainly using cut-based methods. At the start of the Run-2, the physical analysis began to be carried out using Multivariate Analysis (MVA) or Machine Learning (ML) methods. The ML method includes a large number of algorithms, and three of them, which are widely used in high energy physics, were used in this work: boosted decision tree (BDT), multilayer perceptron artificial neural network (TMlpANN), and deep neural network (DNN). The performance of these algorithms was compared. The input variables are the same as those used in the analysis in the ATLAS experiment and are the same for all three ML algorithms. About 3 million signal and 11 million background events were used for training. TMlpANN with optimal hyper-parameters outperforms the other two algorithms when using Root's Toolkit for MVA (TMVA). However, moving to more advanced frameworks such as TensorFlow or PyTorch changes the situation in favor of DNN. But in all cases, BDT remains the most efficient algorithm due to its high training speed.
