Broadening the search for new physics at the LHC using BumpNet

Abstract: We will argue that the fraction of the experimental phase space probed so far at the LHC is surprisingly small compared to all the possible final states. Historically, identifying bumps in invariant mass distributions has been a highly effective method for discovering new particles—most famously, the Higgs boson. In this talk, we introduce BumpNet, a convolutional neural network that rapidly and robustly identifies mass bumps in LHC data. By leveraging AI, BumpNet enables us to explore previously unexamined regions of the LHC phase space, opening new avenues in the search for physics beyond the Standard Model.