Deep machine learning for the PANDA software trigger

Deep machine learning methods have been studied for the software trigger of future PANDA experiment at FAIR, using Monte Carlo simulated data from GEANT-based detector simulation framework PandaRoot. Ten physics channels that cover main topics, including electromagnetic, exotic, charmonium, open charm, and baryonic reaction channels, investigated four different anti-proton beam momenta. Binary multi-class classification together with seven network architectures studied. Finally a residual convolutional neural blocks in binary scheme has chosen due to its extendability, performance stability. The presented study represents feasibility completely software-based system. Compared conventional selection method, deep approach achieved significant efficiency gain up 200\%, while keeping background reduction factor required level 1/1000. Furthermore, it is shown use additional input variables can improve quality subsequent analysis. This shows benefit greatly methods.