Find a way to use a Keras-based model description to drive an LBANN training run. This can occur in one of several ways:

1. Directly support the Keras backend API and use LBANN directly. Note that this is not straight-forward because of the parallel nature (MPI) of LBANN.

2. Create a scripted interface that can take Keras code and specify and launch the same model in LBANN.