Fast Convergence of Natural Gradient Descent for Overparameterized Neural Networks
share
Natural gradient descent has proven effective at mitigating the effects of pathological curvature in neural network optimization, but little is known theoretically about its convergence properties, especially for nonlinear networks. In this work, we analyze for the first time the speed of convergence for natural gradient descent on nonlinear neural networks with the squared-error loss. We identify two conditions which guarantee the efficient convergence from random initializations: (1) the Jacobian matrix (of network's output for all training cases with respect to the parameters) is full row rank, and (2) the Jacobian matrix is stable for small perturbations around the initialization. For two-layer ReLU neural networks (i.e., with one hidden layer), we prove that these two conditions do in fact hold throughout the training, under the assumptions of nondegenerate inputs and overparameterization. We further extend our analysis to more general loss functions. Lastly, we show that K-FAC, an approximate natural gradient descent method, also converges to global minima under the same assumptions.
READ FULL TEXT