pLISA: a parallel Library for Identification and Study of Astroparticles

INFN has produced a Machine Learning library in Python that applies Convolutional Neural Networks to various common problems in the field of astroparticle identification and study in suitable detectors. The library itself makes few assumptions and has few requirements that are easily met in most astroparticle detectors. The Parallel Library for Identification and Study of Astroparticles (pLISA) has been tested against simulated events for the KM3NeT/ARCA detector. Interesting preliminary results have been obtained for up/down-going particle classification, muon/electron neutrino classification, Z component of the direction and energy estimation. Already with very little optimization work and using limited hardware resources (one NVidia GTX GPU), pLISA was shown to compete with traditional algorithms. The approach allows improvements and also portability to other detectors. pLISA is based on commonly used open source frameworks, which helps ensuring portability and scalability.