Instead of using finite differences to approximate derivatives, which demands oversampling to minimize numerical errors, a convolutional neural network ... Maxwell's equations, which describe ...