A Variational AutoEncoder for Model Independent Searches of New Physics at LHC

We present a feasibility study for the use of a generative, probabilistic model, a Variational Autoencoder (VAE), to detect deviations from Standard Model (SM) physics in an electroweak process at the Large Hadron Collider (LHC). The new physics responsible for the anomalies is described through an Effective Field Theory (EFT) approach: the SM Lagrangian is Taylor-expanded and the higher order terms cause deviations in the kinematic distributions of the observables, and are thus identified by the model as anomalous contributions with respect to SM. Since the training of the model involves almost only SM events, the proposed strategy is largely independent from any assumption on the nature of the new physics signature. To test the proposed strategy we use parton level generations of Vector Boson Scattering (VBS) events at the LHC, assuming an integrated luminosity of 350 fb ^{- 1}.