Speaker
Description
Precision measurements at LHC and HL-LHC require reconstruction of the full event kinematics to derive the observables of interest, a task complicated by two distinct source of ambiguity: the kinematics of neutrinos, which escape the detector undetected, and parton origin of the observed final states. We present VyPER, a Graph Neural Network that provides comprehensive event reconstruction with generative graph diffusion and hypergraph representation learning. It is designed to simultaneously reconstruct neutrino kinematics and identify the partonic origin of the final states, offering a unified alternative to traditional multi-step reconstruction pipelines. We present performance studies comparing VyPER against existing state-of-the-art approaches across a range of physics processes, and discuss its prospective role in precision top-quark and electroweak measurements at ATLAS and CMS.