[dapperdrake]

Where do Hessians come into play for neural networks? It seems like they just use auto-diff to compute the Jacobian or the gradient for backpropagation.

The theoretical results sometime look at the second order derivative.