Simulation-assisted decorrelation for resonant anomaly detection

A growing number of weak- and unsupervised machine learning approaches to anomaly detection are being proposed significantly extend the search program at Large Hadron Collider elsewhere. One prototypical examples for these methods is resonant new physics, where a bump hunt can be performed in an invariant mass spectrum. significant challenge that rely entirely on data they susceptible sculpting artificial bumps from dependence classifier mass. We explore two solutions this by minimally incorporating simulation into learning. In particular, we study robustness Simulation Assisted Likelihood-free Anomaly Detection (SALAD) correlations between Next, propose approach only uses decorrelation but Classification without Labels (CWoLa) achieving signal sensitivity. Both compared using full background fit analysis simulated LHC Olympics robust data.