One of the most well-known astrophysical sources of electron antineutrinos is the diffuse supernova neutrino background (DSNB), composed of neutrinos emitted from all past core-collapse supernovae. Since the predicted DSNB flux depends on astrophysical parameters such as the supernova rate and the cosmic star formation history, its detection is expected to provide complementary information to optical observations and help refine these models. Other potential sources of astrophysical neutrinos include solar antineutrinos---indicating physics beyond the Standard Model---as well as antineutrinos from light dark matter annihilation and Hawking radiation from primordial black holes. KamLAND, a 1-kiloton liquid scintillator detector, observes electron antineutrinos via inverse beta decay using a delayed-coincidence technique. Owing to its high sensitivity in the 10 MeV energy region, KamLAND has a unique advantage in the search for astrophysical neutrinos. The dominant background in this energy range arises from neutral-current interactions of atmospheric neutrinos. To reduce this background, we are developing a deep neural network-based classification method that leverages differences in the spatiotemporal distribution of photomultiplier tube hit patterns, caused by differences in interaction topology. Our model achieves a background rejection efficiency exceeding 65% while maintaining a signal acceptance above 75%. The systematic uncertainty introduced by the deep neural network is suppressed to approximately 10%. We plan to apply this event selection method to the full KamLAND dataset in the near future.
M. Eizuka (Wed,) studied this question.