JEDI-net: a jet identification algorithm based on interaction networks

Abstract

We investigate the performance of a jet identification algorithm based on interaction networks (JEDInet) to identify all-hadronic decays of high-momentum heavy particles produced at the LHC and distinguish them from ordinary jets originating from the hadronization of quarks and gluons. The jet dynamics are described as a set of one-to-one interactions between the jet constituents. Based on a representation learned from these interactions, the jet is associated to one of the considered categories. Unlike other architectures, the JEDI-net models achieve their performance without special handling of the sparse input jet representation, extensive pre-processing, particle ordering, or specific assumptions regarding the underlying detector geometry. The presented models give better results with less model parameters, offering interesting prospects for LHC applications.

Exploring anomalous couplings in Higgs boson pair production through shape analysis

“…While the benchmark points representing a cluster are isolated points in the parameter space, the procedure we propose here allows us to associate certain shapes more straightforwardly with distinct regions in the parameter space. The application of machine learning techniques in high energy physics, in particular to constrain the EFT/new physics parameter space, has been brought forward already some time ago [68][69][70][71], with successful applications in jet and top quark identification [72][73][74][75][76][77][78][79][80][81], new physics searches [70,71,[82][83][84][85][86][87][88][89][90] and PDFs [91]. Shape analysis with machine learning has been applied already to constrain anomalous Higgs-vector boson couplings in HZ production [92].…”

Interaction networks for the identification of boosted H→bb¯ decays

“…In Ref. [47], it was shown that the area under the receiver operating characteristic (ROC) curve (AUC), accuracy, and background rejection at a 30% true positive rate (TPR) of a simple IN architecture trained with the same dataset is within 1%, 0.5%, and 40% of those of ParticleNet, while using 70% fewer parameters. Unsupervised, semisupervised, and weakly supervised methods have also been proposed, mainly to tag t jets or jets coming from postulated new particles [74][75][76][77][78][79][80][81][82][83][84].…”

“…Graph networks [47,71,73,93] and the related particle flow networks [94] have recently been used for other kinds of jet tagging, matching or exceeding the performances of other DL approaches, for event classification [95,96], for charged particle tracking in a silicon detector [97,98], for mitigation of the effects pileup [99], and for particle reconstruction in irregular calorimeters [98,[100][101][102] and the IceCube experiment [96].…”

“…Moreover, Ref. [47] showed that the learned features of an IN correlate with known jet substructure variables. It was further demonstrated that the IN architecture outperformed other deep neural networks (DNNs), such as dense, convolutional, and recurrent networks, for a jet-substructure classification task.…”

Object condensation: one-stage grid-free multi-object reconstruction in physics detectors, graph, and image data

“…Neural network based algorithms offer the possibility of retaining the information, and furthermore, there is a trend towards employing such algorithms for more tasks in high energy physics further towards the beginning of the reconstruction sequence. In this context, graph neural networks [32] are receiving increasing attention because they allow direct processing of detector inputs or particles, which are both sparse and irregular in structure [33][34][35]. However, when attempting to also incorporate the seeding step together with subsequent steps, the above mentioned methods from computer vision are not directly applicable.…”