Improving application of Bayesian Neural Networks to discriminate neutrino events from backgrounds in reactor neutrino experiments

Abstract

The application of Bayesian Neural Networks (BNN) to discriminateneutrino events from backgrounds in reactor neutrino experiments hasbeen described in ref. [1]. In the paper, BNN are also usedto identify neutrino events in reactor neutrino experiments, but thenumbers of photoelectrons received by PMTs are used as inputs to BNNin the paper, not the reconstructed energy and position of events.The samples of neutrino events and three major backgrounds from theMonte-Carlo simulation of a toy detector are generated in the signalregion. Compared to the BNN method in [1], more8He/9Li background and uncorrelated background in thesignal region can be rejected by the BNN method in the paper, butmore fast neutron background events in the signal region areunidentified using the BNN method in the paper. The uncorrelatedbackground to signal ratio and the 8He/9Li background tosignal ratio are significantly improved using the BNN method in thepaper in comparison with the BNN method in ref. [1]. But thefast neutron background to signal ratio in the signal region is abit larger than the one in ref. [1].