Classical and Quantum Machine Learning Targeting Small Excesses in ATLAS

ATLAS experiment at the LHC offers rich Beyond The Standard Model searches program, although no significant sign of new physics has been seen to date. However, small anomalies were observed in multiple analyses, while little attention is given to finding the connections of observed small excesses in the data. The MATREX research proposal addresses the connection of small excesses using large scale reinterpretation of 19-parameter Phenomenological Minimal Supersymmetry Model (pMSSM) by the existing Supersymmetry searches in ATLAS. Since typically BSM analyses are optimized using simplified models, their performance is evaluated using realistic pMSSM models. Models to which good sensitivity was obtained by multiple analyses but were not excluded due to an excess in the data, are considered in the study. First, a large number of pMSSM models are generated and interpreted using existing analyses. Second, common features of the model parameters and final states are studied using unsupervised learning and advanced visualization techniques. In addition, common features of data events from selections that have seen an excess are studied using unsupervised learning. Finally, first implementations of this complex problem are studied using quantum machine learning in the quest for the quantum advantage. These technically and computationally demanding tasks will benefit from strong expertise and CPU/GPU/QPU resources provided by the SMASH institutes and CERN laboratory.