Reweighting simulated events using machine-learning techniques in the CMS experiment.

Area of Science:

• High-energy particle physics
• Computational physics
• Machine learning applications

Background:

• Particle physics data analysis requires accurate simulations of particle collisions and detector responses.
• Current simulation methods, particularly detector simulation, are computationally intensive, demanding significant CPU resources.
• Large Hadron Collider (LHC) experiments rely on extensive simulated event samples for data analysis.

Purpose of the Study:

• To introduce and evaluate machine learning (ML) techniques for reweighting simulated particle physics event samples.
• To demonstrate how ML can adapt existing simulated samples to different physics parameters or simulation programs, reducing computational overhead.
• To enhance the efficiency of generating simulated data for LHC experiments, particularly for precision measurements.

Main Methods:

• Utilizing machine learning algorithms to assign weights to simulated events.
• Reweighting a single simulated sample to represent variations in simulation parameters or alternative simulation models.
• Applying the ML reweighting method to simulated top quark pair production events at the LHC.

Main Results:

• Successfully reweighted simulated samples to different model variations and higher-order calculations.
• Demonstrated that ML reweighting effectively incorporates necessary information into a single sample via event weights.
• Validated the ML approach as a viable alternative to repeated, costly detector simulations.

Conclusions:

• ML-based reweighting significantly reduces the computational burden associated with particle physics simulations.
• This method is a crucial component for the future computing strategy of experiments like the Compact Muon Solenoid (CMS).
• The technique will facilitate high-precision measurements at the High-Luminosity LHC by improving simulation efficiency.