Aug 2 – 6, 2021
Europe/Brussels timezone

Mitigation of systematic errors induced biases in ML-based selection

Aug 5, 2021, 10:10 AM


Parallel contribution H. Statistical Methods for Physics Analysis in the XXI Century Parallels Track H


Francisco Matorras (Instituto de Fisica de Cantabria, Santander, IFCA (ES))


One of the main limitations in particle physics analyses with ML-based selection is the understanding of the implications of systematic uncertainties. The usual approach being the training using samples without systematic effects and estimating their contribution to the magnitudes measured on modified test samples. We propose here a method based on data augmentation to incorporate the systematics at the training time, which provides both an improvement in the performance and a reduction in the biases.

Primary authors

Francisco Matorras (Instituto de Fisica de Cantabria, Santander, IFCA (ES)) Dr Pablo Martínez-Ruiz del Árbol (IFCA- Universidad de Cantabria-CSIC) Mr Luis Crespo (Universidad de Cantabria)

Presentation materials