Morphology for jet classification

We introduce a jet tagger based on neural network analyzing the Minkowski functionals (MFs) of pixelated images. The MFs are geometric measures binary images, and they can be regarded as generalization particle multiplicity, which is an important quantity in tagging. Their changes by dilation encode constituents' structures that appear at various angular scales. explicitly show this analysis using considered constrained convolutional (CNN). Conversely, CNN could model limit large network. example decision boundary correlates strongly with value semivisible tagging hidden valley scenario. independent infrared collinear (IRC)-safe observables commonly used physics. combine morphological IRC-safe relation models two-point energy correlations. While resulting uses input parameters, it shows comparable dark top performances to CNN. architecture has significant computational advantages when available data limited. its performance much better than small number training samples. also qualitatively discuss their parton shower dependency. results suggest efficient parametrization IRC-unsafe feature space jets.