Prediction of High Energy Particle Kinematics via Masked Autoencoding

In high energy physics, much research revolves around the study of particles produced by colliding protons at near the speed of light. The Higgs boson which is a heavy and highly unstable elementary particle is not present in nature but is generated in high energy collisions. Because it’s unstable and decays before reaching the edge of the detector, we study it by observing its decay products which tend to form conical jets and then reconstruct the Higgs bosons from these observations. One technique we might use to reconstruct these particles is to pass the kinematic variables (momentum, energy, etc.) to a transformer autoencoder while masking either subsets of kinematics or the kinematics for an entire particle. This could lead to improvements of reconstruction for more difficult to reconstruct particles such as neutrinos.