Liquid argon time projection chamber (LArTPC) is a novel technology for neutrino detection because of its excellent imaging capability of charged particles. However, it is challenging to reconstruct and analyze LArTPC events efficiently in large scale detectors. In this talk, I will review some revolutionary ideas in machine learning, and demonstrate the use of these techniques to tackle the challenge in LArTPC event reconstruction.