A deep neural network for simultaneous estimation of b quark energy and resolution

We describe a method to obtain point and dispersion estimates for the energy of jets arising from bottom quarks (b jets) in proton-proton (pp) collisions at the CERN LHC. The algorithm is trained using a large simulated sample of b jets produced in pp collisions recorded at an energy of $\sqrt{s}=13~\mathrm{TeV}$ and validated on data recorded by the CMS detector in 2017 with an integrated luminosity of $41~\mathrm{fb}^{-1}$. A multivariate regression estimator employing jet composition and structure information and the properties of the associated reconstructed secondary vertices is implemented using a deep feed-forward neural network. The results of the algorithm are used to improve the experimental sensitivity of analyses that make use of b jets in the final state, such as the recently published observation of the Higgs boson decay to a bottom quark-antiquark pair.