Searches for heavy neutral leptons with machine learning at the CMS experiment

TL;DR

This paper reports on two CMS searches for heavy neutral leptons using machine learning to improve signal-background separation, setting new exclusion limits across a wide mass range.

Contribution

It introduces machine learning techniques into HNL searches at CMS, covering both prompt and displaced decay signatures and extending exclusion limits.

Findings

Set new exclusion limits on HNL coupling strength.

Applied machine learning to enhance background discrimination.

Achieved results surpassing previous experimental limits.

Abstract

Two recent searches for heavy neutral leptons (HNLs) performed with proton-proton collision data recorded at 13 TeV by the CMS experiment are presented. A prompt search in the trilepton final state analyses events with exactly three charged leptons originating from the primary proton-proton interaction vertex, targeting HNL masses between 10 GeV and 1.5 TeV. A displaced search in the dilepton final state analyses events with exactly one prompt charged lepton and a second nonprompt charged lepton associated with a jet and a secondary vertex, targeting HNL masses between 1 and 20 GeV. In both searches, machine-learning methods are applied to separate the HNL signal from the standard model background. Exclusion limits are set on the HNL coupling strength as a function of the HNL mass, covering different mass ranges and HNL scenarios. In several cases, the results exceed previous limits.