Exploring Higgs bosons at high energies: From jets as graphs to fast machine learning on FPGAs

Since the discovery of the Higgs boson at the LHC, we have strived to measure its properties and interactions to identify any deviations from their expectations that may give hints for new laws of physics. These deviations may be amplified at high energies. When a highly energetic Higgs boson is produced, its decay products form collimated showers, called jets, that overlap within our detectors, making it challenging to reconstruct and identify them using traditional techniques. Instead, we have turned to machine learning methods to enable measuring highly energetic Higgs bosons. In this talk, I will describe two significant machine-learning developments that have had major impacts on Higgs boson physics at the LHC. The first is treating jets as graphs and analyzing the pairwise relationships between particles in "graph neural networks" to achieve better signal-to-background discrimination. The second is accelerating machine learning algorithms for the trigger so they can run in nanoseconds on field-programmable gate arrays to preserve precious signal events that would otherwise be discarded forever.