The Standard Model of Particle Physics has been extremely successful in explaining a myriad of phenomena. That said, the model must be incomplete since it does not explain key questions such as "what is dark matter?", "why is the universe made of matter and not antimatter?", etc.
This research group uses data from the CMS experiment at the CERN Large Hadron Collider to search for evidence of new physics, beyond the Standard Model. Their focus is on new physics signatures that are very challenging to distinguish from the known backgrounds; for example, they target signatures of top quark pair production in final states with large jet multiplicity, rather than a large imbalance of momentum.
Because of the challenging target, the creative use of machine learning is essential. The group is exploring the use of deep neural networks with custom loss functions, as well as adversarial neural networks, to design a physics analysis that is both highly sensitive and robust against systematic uncertainties.
Machine learning techniques such as graph neural networks are also used for event reconstruction for high energy physics. A new focus for this group is to study the robustness of these networks against possible mismodeling of the underlying physics in their simulations, to evaluate methods to improve this robustness, and to explore pretraining on simulation and fine-tuning on early physics data.