Performance of the DeepJet b tagging algorithm using 41.9/fb of data from proton-proton collisions at 13TeV with Phase 1 CMS detector

The identification of jets originating from b quarks (b jets) is of great importance to many physics analyses. Multivariate analysis techniques have been traditionally used in flavour tagging algorithms, however there has been a recent step change towards using deep learning algorithms due to their suitability to complex multi-classification problems. We present here the first measurement of the performance of the DeepJet algorithm using 41.9/fb of data collected from proton-proton collision at 13 TeV using the CMS detector. We derive efficiency scale factors to correct for the difference in performance on 2017 Monte Carlo simulation and data using two performance measurement techniques. We also show first studies on the dependence of the performance of DeepJet on the size of the training sample and its stability with respect to the random weight initialisation.