Machine learning strategies for LHC data analysis
The LHC is the most powerful particle accelerator operating today. Colliding protons with a center-of-mass energy of 13 TeV, it is sensitive to physics at the shortest distance scales that we can currently access experimentally. It has the potential to provide answers to some of the open questions of particle physics or reveal new surprises. The random nature of the collision processes and the complexity of the final states (which typically include hundreds of particles) necessitate a significant data analysis effort to extract the underlying information. The most straightforward analysis methods are already well-established and have produced a variety of important results, including the discovery of the Higgs boson. To extract further information, one needs to dig deeper in the data. One of the most natural tools for dealing with complex structures in large amounts of data is machine learning and we have been pursuing several projects in this direction.
In one recent project we proposed several machine learning strategies to search for new physics signals in wavelet transforms of kinematic distributions.
Searching for periodic signals in kinematic distributions using continuous wavelet transforms
Hugues Beauchesne, Yevgeny Kats
Eur. Phys. J. C 80 (2020) 192; arXiv:1907.03676 [hep-ph]
Another work in progress involves the application of machine learning techniques to identify, in a largely model-independent fashion, anomalous jets whose origin might not be a quark or gluon but physics beyond the Standard Model.