Enhanced Higgs boson to τ(+)τ(-) search with deep learning.

Area of Science:

• Particle Physics
• High Energy Physics
• Machine Learning in Physics

Background:

• The Higgs boson is theorized to impart mass to fundamental fermions.
• Current Large Hadron Collider (LHC) analyses require more data to reach the 5σ significance threshold for Higgs boson detection.
• Advanced analysis techniques are needed to improve statistical power.

Purpose of the Study:

• To apply deep learning techniques for enhanced detection of Higgs boson decays to tau leptons.
• To improve the statistical power of Higgs boson analysis at the LHC.
• To explore the efficacy of deep neural networks in particle physics data analysis.

Main Methods:

• Utilized deep neural networks to identify Higgs boson decays into tau lepton pairs.
• Employed a Bayesian optimization algorithm for hyperparameter tuning of network architecture and training.
• Developed a deep network with eight nonlinear processing layers.

Main Results:

• The deep neural network classifier outperformed shallow classifiers.
• The method improved discovery significance, equivalent to a 25% increase in data.
• The network learned complex data representations without physicist-engineered features.

Conclusions:

• Deep learning offers a powerful approach to enhance statistical significance in particle physics.
• The developed deep neural network architecture effectively detects Higgs boson decays.
• This methodology can accelerate discoveries in high energy physics by maximizing existing data.