Choose timezone
Your profile timezone:
University of Kansas
PPP Theory Seminar: Sung Hak Lim (Rutgers University), "Morphology for Jet Classification"
→
US/Central
Description
Meeting link: https://kansas.zoom.us/j/92638240546
Meeting ID: 926 3824 0546
Passcode: 779046
Abstract:
Minkowski functionals in integral geometry are useful mathematical tools for describing geometric features of point clouds and have been applied in cosmology and material sciences. In this talk, we introduce a morphological analysis for jet classification based on machine learning on the Minkowski Functionals (MFs) of a jet image. The MFs systematically provide jet constituent's geometric information, which is complementary to the IRC safe observables commonly used in jet physics. We use these MFs together with a relation network based on IRC safe two-point energy correlations to model a new classifier and show what kind of substructure variables are covered by this model. The tagging performance of this model is comparable to the convolutional neural network trained on jet images, even though the new classifier is computationally cheaper.
The result suggests that the MFs can be an efficient parameterization of the feature space of jets.
