2 code implementations • 7 Feb 2017 • Moshe Looks, Marcello Herreshoff, DeLesley Hutchins, Peter Norvig
However, since the computation graph has a different shape and size for every input, such networks do not directly support batched training or inference.
