LPC Physics Forum

Thursday 11 Dec 2025, 06:00 → 14:00 US/Central

Sunrise (WH11NW) (FNAL/Zoom)

https://cern.zoom.us/j/96898909340

13:00 → 14:00 Enhanced reconstruction of dileptonic top quark-antiquark events using supervised machine learning methods

A machine learning based reconstruction of the top quark-antiquark kinematic system in the dileptonic decay channel is presented. The presence of two neutrinos poses a significant challenge to top quarks reconstruction in this channel. A multilayer perceptron and a transformer model are trained on simulated events to estimate the four-momenta of both top quarks in each event. The transformer achieves an average of 30% improvement in resolution of top quark and $t\bar{t}$ kinematic variables compared to the commonly used analytical method, and outperforms the multilayer perceptron model by about 7%. Additionally, It is able to reconstruct approximately 5% of events that the analytical method fails to reconstruct. This work offers a new approach for precision top quark measurements and searches for new physics in the dileptonic channel.

Speaker: Yao Yao (Purdue University (US))