Advancing Higgsino Searches by Integrating ML for Boosted Object Tagging and Event Selection

TL;DR

This paper introduces a new search strategy for Higgsinos at the LHC that uses graph neural networks and boosted decision trees to better identify boosted objects, significantly improving detection sensitivity.

Contribution

The work integrates graph neural networks with traditional classifiers to enhance boosted object tagging and event selection in Higgsino searches, representing a novel approach.

Findings

GNNs improve fat jet characterization accuracy.

Enhanced signal-background discrimination with combined GNN and BDT.

Significant sensitivity gains in Higgsino detection at the LHC.

Abstract

Higgsinos near the TeV mass range are highly motivated as they offer an elegant solution to the naturalness problem in the Standard Model. Extensive searches for such higgsinos within the framework of General Gauge Mediation (GGM) have been conducted by both the ATLAS and CMS collaborations. However, the sensitivity of these searches in the hadronic channel remains limited, primarily due to the reliance on traditional substructure-based techniques for fat jet identification. In this work, we present a novel search strategy that leverages graph neural networks (GNNs) to improve the characterization of fat jets originating from W/Z/h bosons, top quarks, and QCD-initiated light quarks and gluons. The GNN scores, combined with a boosted decision tree (BDT) classifier, enhance signal-background discrimination, offering a significant improvement in sensitivity for higgsino searches at the LHC.