Karlsruhe 2024 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
T: Fachverband Teilchenphysik
T 46: Di-Higgs 1 (bbττ)
T 46.1: Talk
Tuesday, March 5, 2024, 16:00–16:15, Geb. 30.41: HS 1
Application of Deep Sets Neural Networks for the bbτ−τ+ Di-Higgs Analysis with the CMS Experiment — •Stella Felice Schaefer, Peter Schleper, Philip Daniel Keicher, and Bogdan Wiederspan — University of Hamburg, Hamburg, Germany
Machine Learning has found a wide range of applications in particle physics. In the context of the CMS bbτ−τ+ Di-Higgs Analysis neural networks are tasked with the classification of signal and background processes in order to measure the Di-Higgs coupling constant κλ as well as the coupling of two Higgs bosons to two vector bosons κ2V.
Common problems for the proper application of neural networks are the variable number of jets per event, requiring padding of empty jets for a fixed number of input features, as well as the input order of jets passed to the network. Deep Sets neural networks provide a solution to both aforementioned problems, as these networks don’t require fixed input shapes and act permutation invariant on the input.
This study aims to thoroughly test Deep Sets neural networks in the context of the CMS bbτ−τ+ Di-Higgs Analysis for the application of signal process classification as well as signal vs. background classification and compares the performance to standard feed-forward architectures.
Keywords: Machine Learning; Deep Sets; Di-Higgs