Erlangen 2026 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

T: Fachverband Teilchenphysik

T 28: Data, AI, Computing, Electronics III

T 28.7: Talk

Tuesday, March 17, 2026, 17:45–18:00, KH 00.024

Reconstructing missing transverse momentum for electroweak precision measurements at the ATLAS experiment — •Gabriel Sanchez Shestakova, Matthias Schott, Timo Saala, and Philip Bechtle — Physikalisches Institut, Bonn, Germany

We present first studies of a supervised Machine Learning (ML) approach for reconstructing missing transverse energy (MET) using Particle Flow (PF) information acquired from open Large Hadron Collider (LHC) data. Using simulated W⁺ → µ⁺ ν events at √s = 8 TeV, we compute an event-level MET observable directly from PF objects by explicitly reconstructing transverse momentum components and forming an analytically calculated MET, which serves as a controlled regression target for the ML approach.

A fully connected multilayer perceptron (MLP) is trained on per-event PF momentum components, with variable-length PF collections handled via zero-padding to a fixed input dimension. We also develop a graph neural network (GNN) approach that operates on variable-size PF representation. Quantile-based selections on the number of PF objects and on the calculated MET are applied prior to training in order to mitigate outliers and reduce input distribution mismatches.

Keywords: missing transverse energy; machine learning