Performance of boosted decision trees for combining ATLAS b-tagging methods

This note evaluates the performance of boosted decision trees for combining the information from different ATLAS b-tagging algorithms into a single jet classifier. The rejection of light quarks given by boosted decision trees is estimated using a Monte Carlo simulation of $WH$ and $t\bar{t}$ events. It is shown that this approach yields significant gains in the rejection of light quarks with respect to a tagging algorithm based in 3D impact parameters and reconstructed secondary vertices.