A jet tagging algorithm of graph network with HaarPooling message passing

Fei Ma, Feiyi Liu, Wei Li

Submitted on 25 October 2022, last revised on 14 November 2022


Recently methods of graph neural networks (GNNs) have been applied to solving the problems in high energy physics (HEP) and have shown its great potential for quark-gluon tagging with graph representation of jet events. In this paper, we introduce an approach of GNNs combined with a HaarPooling operation to analyze the events, called HaarPooling Message Passing neural network (HMPNet). In HMPNet, HaarPooling not only extract the features of graph, but also embed additional information obtained by clustering of k-means of different particle observables. We construct Haarpooling from three different observables: absolute energy logE, transverse momentum logpT and relative coordinates (Δη,Δϕ), then discuss their impacts on the tagging and compare the results with those obtained via message passing neutral network (MPNN) and ParticleNet (PN). The results show that an appropriate selection of information for HaarPooling enhance the accuracy of quark-gluon tagging, for adding extra information of logPT to the HMPNet outperforms all the others, meanwhile adding relative coordinates information (Δη,Δϕ) is not very beneficial.


Subjects: High Energy Physics - Experiment; Computer Science - Computer Vision and Pattern Recognition; Computer Science - Machine Learning; High Energy Physics - Phenomenology


An event graph with node and edge weights for a specific simulated event of the process $pp \to Z / \gamma^{\ast} + j + X \to \mu^{+} \mu^{-} + j + X $.