Weakly-Supervised Anomaly Detection With Conditional VAEs

Machine learning-based anomaly detection techniques offer exciting possibilities to significantly extend the search for new physics at the Large Hadron Collider (LHC) and elsewhere by reducing the model dependence. In this work, we focus on resonant anomaly detection, where we train a Variational Autoencoder in background regions and interpolated into a signal region to provide an estimate of the Standard Model background. This estimate can then be compared with data using a machine learning classifier. We demonstrate this idea by conducting a di-jet resonance search using the LHC Olympics 2020 challenge dataset. Anomaly detection methods have already been used in a few searches at the LHC, but they are still at the very early stage of their development and this work will help to push forward their use in the LHC. Anomaly detection methods using ML augment the presence of potential signals by using other features other than mass. In this work the VAE is conditioned on the di-jet invariant mass and uses six other kinematic variables to generate background events in signal regions. The preliminary results are promising and will help in future LHC anomaly detection searches.