Parts | Days | Selection | Search | Updates | Downloads | Help

T: Fachverband Teilchenphysik

T 88: Top Physics IV

T 88.4: Talk

Friday, March 20, 2026, 09:45–10:00, KH 00.011

Is Assignment All You Need? — •Siemen Aulich, Katharina Behr, and Eleanor Jones — DESY, Hamburg, Germany

Many of the recent highlights of particle physics research are related to top quark physics’. These include both the tests of spin correlations and quantum effects in pairs of top quarks (tt), and the observation of a possible quasi-bound state resonance in the tt invariant mass spectrum. Both effects are predominantly studied in dilepton decays in a mass range close to the production threshold.

Probing this system requires a precise reconstruction of the top quarks, which is complicated by the presence of the two neutrinos. While analytical regression strategies primarily focus on inferring the neutrino momenta, the problem of correctly assigning b-jets to their parent top quarks remains largely unstudied. However, many of the sensitive variables used in tt precision measurements depend critically on the correct assignment of the jets. Inspired by the success of machine learning architectures in tackling the assignment challenge for hadronic decay channels, this work investigates using a transformer model for the dilepton channel. An architecture specifically tailored to this channels topology is shown to outperform all existing methods. Furthermore, it is investigated how these efforts can be combined with neutrino regression methods to offer a full reconstruction pipeline. Applying these methods to existing and upcoming analyses promises further enhancements in the sensitivity and precision of searches and measurements alike.

Keywords: Dilepton; Machine Learning; Transformer