signature=5bb535932c87e84edcb2830f5cdf79a4,Search for rare SM processes in the ET +b-jets signature ...

摘要：

We train our network using the pre-tag data sample weighted by the Tag-Rate-Matrix to model MJ events. This choice is preferred over the use of a Monte Carlo QCD background because it allows not only to reject QCD but also a part of the other electroweak backgrounds. With this technique, it is possible to reject about 90% (70%) of the MJ (overall) background while only loosing 10% of the signal. The signal sample is composed of Monte Carlo events in proportion to the relative size of each signal component.Figure 1a shows the difference in shape for some of the network input variables. Each of the variables is validated against data, as shown in Figure 1b. In addition to the region where the network is derived, we use five other control regions in which we check both the inputs to the network and its output. This makes us confident in using multivariate analysis to find a signal.

展开