Sobolev Acceleration and Statistical Optimality for Learning Elliptic Equations via Gradient Descent

Neural Information Processing Systems 

We apply gradient descent to a general class of objective functions for the reconstruction.