A Diffusion Approximation Theory of Momentum SGD in Nonconvex Optimization
This provides theoretical insights for researchers and practitioners using MSGD in machine learning, though it is incremental as it builds on existing momentum methods.
The paper tackled the lack of theoretical understanding of Momentum SGD (MSGD) in nonconvex optimization by analyzing its behavior using diffusion approximations, showing that momentum helps escape saddle points but hinders convergence near optima without annealing.
Momentum Stochastic Gradient Descent (MSGD) algorithm has been widely applied to many nonconvex optimization problems in machine learning, e.g., training deep neural networks, variational Bayesian inference, and etc. Despite its empirical success, there is still a lack of theoretical understanding of convergence properties of MSGD. To fill this gap, we propose to analyze the algorithmic behavior of MSGD by diffusion approximations for nonconvex optimization problems with strict saddle points and isolated local optima. Our study shows that the momentum helps escape from saddle points, but hurts the convergence within the neighborhood of optima (if without the step size annealing or momentum annealing). Our theoretical discovery partially corroborates the empirical success of MSGD in training deep neural networks.