6–10 Jul 2026
University of the Western Cape
Africa/Johannesburg timezone
**Tours now open!** Registration is now closed - All registration payments are due before 23:39 SAST on 26 June.

Guided Weakly Supervised Machine Learning for Resonance Searches in Low-Energy Regions of Phase Space in the ${\gamma\gamma + 1\tau}$ channel at the Large Hadron Collider

7 Jul 2026, 09:30
20m
Lecture Hall GH3 (University of the Western Cape)

Lecture Hall GH3

University of the Western Cape

Oral Presentation Track B - Nuclear, Particle and Radiation Physics Nuclear, Particle and Radiation Physics -2

Speaker

Phodiso Maroeshe (School of Physics and Institute for Collider Particle Physics, University of the Witwatersrand, Johannesburg, Wits 2050, South Africa)

Description

Searches for new resonances at hadron colliders are typically optimized for benchmark models and high transverse momentum final states in order to suppress large Standard Model backgrounds. However, potential signals of new physics may appear in less explored regions of phase space where traditional analyses may have reduced sensitivity. This project investigates the use weak supervision in resonant searches on the $\gamma \gamma + 1\tau$ final state in a phase space of reduced hadronic activity, with a lowered jet jet $p_T$ threshold of $10 \: GeV$, as opposed to a space of high $p_T$ objects. To identify potential signals without relying on event-level truth labels, a guided weakly supervised machine learning approach based on the is employed. In this method, a classifier is trained to distinguish between background and a mixed sample of signal and background, allowing discriminating features of potential new physics signals to be learned. A Graph Neural Network (GNN) is employed as the deep learning model, as its object-based representation and permutation-invariant architecture naturally accommodate the variable jet multiplicity and complex event topology characteristic of hadron collider final states. Crucially, the guidance for weak label construction is derived from the di-photon invariant mass regions, defined by a signal window of $144$-$156 \: GeV$ and sidebands spanning $132$-$168 \: GeV$, excluding the signal region. This study aims to assess the potential of weak supervision techniques to improve sensitivity to subtle signals that may reside in previously under-explored regions of phase space in collider data.

Apply for student award at which level: MSc
Consent on use of personal information: Abstract Submission Yes, I ACCEPT

Author

Phodiso Maroeshe (School of Physics and Institute for Collider Particle Physics, University of the Witwatersrand, Johannesburg, Wits 2050, South Africa)

Co-authors

Bruce Mellado (Institute of High Energy Physics, Beijing, University of Chinese Academy of Sciences, 19B Yuquan Road, Shijingshan District, Beijing, China) Edward Nkadimeng (School of Physics and Institute for Collider Particle Physics, University of the Witwatersrand, Johannesburg, Wits 2050, South Africa) Dr Mukesh Kumar (University of the Witwatersrand)

Presentation materials