Heidelberg 2022 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
T: Fachverband Teilchenphysik
T 26: Data Analysis, Information Technology and Artificial Intelligence
T 26.7: Talk
Monday, March 21, 2022, 17:45–18:00, T-H39
Regression of Missing Transverse Momentum (MET) with Graph Neural Networks — •Nikita Shadskiy1, Matteo Cremonesi2, Jost von den Driesch1, Lindsey Gray3, Ulrich Husemann1, Yihui Lai4, and Michael Waßmer1 — 1Institute of Experimental Particle Physics (ETP), Karlsruhe Institute of Technology (KIT) — 2University of Notre Dame — 3Fermilab — 4University of Maryland
Neutral particles that are only interacting weakly, like neutrinos, which are known from the standard model, or other, still unknown, particles in theories beyond the standard model, can be measured indirectly using the missing transverse momentum (MET). Analyses which search for specific invisible particles or expect such particles in their final state need well reconstructed MET. The reconstruction of MET is sensitive to e.g. experimental resolutions, mismeasurements of reconstructed particles or pileup interactions and is therefore a challenging task.
This talk will give an overview about a new approach to reconstruct MET with graph neural networks using information from particle flow (PF) candidates. Particle flow is an algorithm used by the CMS collaboration to reconstruct particles by combining information from different detector parts. Using graphs is a more intuitive way to describe the topology of an event because it has the advantage to be permutation invariant. Thus, the order of the PF candidates is irrelevant. The graph neural network is optimized to predict a weight for each PF candidate. These predictions are then used to weight the contribution of each PF candidate in the calculation of MET.