Traditional approaches to deep learning can be tedious and time consuming, due to the need for massive amounts of data that needs to be retrained over and over again. Moreover, data is often not available online or is confidential to one organization, so it cannot be combined to create massive AI systems.