Deep learning library for the HPE Cognitive Computing Toolkit (CCT). The NN library layers a DifferentiableField type on top of regular CCT fields. A DifferentiableField behaves like a normal field in most contexts, but includes additional state that allow it to carry gradients. Operators on DifferentiableField types include both forward and backward functions. The NN toolkit can propagate gradients from a loss function back through a directed acyclic graph of differentiable fields and operators.

If you're new to CCT, start with the tutorial.