Backpropagation through structure
From Wikipedia, the free encyclopedia
Backpropagation through structure (BPTS) is a gradient-based technique for training recursive neural networks, proposed in a 1996 paper written by Christoph Goller and Andreas Küchler.[1]
