Accurate Surrogate Amplitudes with Calibrated Uncertainties

TL;DR

This paper introduces a method for training neural surrogates that accurately predict loop amplitudes with well-calibrated uncertainties, enhancing reliability for LHC physics applications.

Contribution

It presents a systematic testing framework for activation functions and a novel approach to disentangle and calibrate systematic and statistical uncertainties in neural surrogates.

Findings

Neural surrogates can predict amplitudes with calibrated uncertainties.

Activation functions are systematically evaluated using Kolmogorov-Arnold Networks.

Disentangling systematic and statistical uncertainties improves reliability.

Abstract

Neural networks for LHC physics have to be accurate, reliable, and controlled. Using neural surrogates for the prediction of loop amplitudes as a use case, we first show how activation functions are systematically tested with Kolmogorov-Arnold Networks. Then, we train neural surrogates to simultaneously predict the target amplitude and an uncertainty for the prediction. We disentangle systematic uncertainties, learned by a well-defined likelihood loss, from statistical uncertainties, which require the introduction of Bayesian neural networks or repulsive ensembles. We test the coverage of the learned uncertainties using pull distributions to quantify the calibration of cutting-edge neural surrogates.