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

T: Fachverband Teilchenphysik

T 33: Higgs II

T 33.10: Talk

Tuesday, March 20, 2018, 18:45–19:00, Z6 - HS 0.004

Verification of Deep Learning Methods for ttH(H→bb) with the CMS Experiment — •Yannik Rath, Martin Erdmann, Benjamin Fischer, Erik Geiser, Dennis Noll, Marcel Rieger, and David Schmidt — III. Physikalisches Institut A, RWTH Aachen University

The analysis of top-quark associated Higgs production allows for a direct measurement of the top-Higgs Yukawa coupling. In the ttH(H→bb) channel, one of the main challenges is the separation of background events, in particular the irreducible background of tt+bb.

In our analysis, we make use of deep neural networks in order to categorize events into the different underlying physics processes. The aim is to separate all processes to improve the simultaneous constraint of both the signal and the background.

While neural networks are a natural choice for this kind of multi-classification, it is often difficult to understand what happens internally in a neural network. This concerns both the interpretation of what information is used by the network and the verification of the results. In this talk, we discuss these questions in the context of our ttH analysis.