(Machine) Learning amplitudes for faster event generation

TL;DR

This paper demonstrates that machine learning regressors can significantly accelerate the evaluation of complex amplitudes in Monte Carlo event generators, maintaining high accuracy and drastically reducing computation time.

Contribution

It introduces a novel approach of replacing exact amplitudes with trained ML regressors to speed up event generation in particle physics simulations.

Findings

ML regressors predict differential distributions with errors below 0.1%.

Prediction times are about 1000 times faster than exact evaluations.

Training takes approximately 7 minutes with small model sizes.

Abstract

We propose to replace the exact amplitudes used in MC event generators for trained Machine Learning regressors, with the aim of speeding up the evaluation of {\it slow} amplitudes. As a proof of concept, we study the process $gg \to ZZ$ whose LO amplitude is loop induced. We show that gradient boosting machines like $\texttt{XGBoost}$ can predict the fully differential distributions with errors below $0.1 \%$, and with prediction times $\mathcal{O}(10^3)$ faster than the evaluation of the exact function. This is achieved with training times $\sim 7$ minutes and regressors of size $\lesssim 30$~Mb. These results suggest a possible new avenue to speed up MC event generators.