Stuart Dreyfus in his paper, “The numerical solution of variational problems” shows a backpropagation model that uses simple derivative chain rule, instead of dynamic programming which earlier backpropagation models were using. This is yet another small step that strengthens the future of deep learning.