Overparameterized Nonlinear Learning: Gradient Descent Takes the Shortest Path?

TL;DR

This paper investigates the behavior of gradient descent in overparameterized nonlinear models, revealing that it converges rapidly to a global minimum close to the initial point along a shortest path, with implications for neural network training.

Contribution

The paper introduces a new theoretical framework showing gradient descent converges efficiently to a near-initial global minimum in overparameterized settings, with novel potential functions and martingale techniques.

Findings

Gradient descent converges geometrically to a global minimum.

Iterates follow a near shortest path from initialization to the solution.

Results apply across various domains like matrix recovery and neural networks.

Abstract

Many modern learning tasks involve fitting nonlinear models to data which are trained in an overparameterized regime where the parameters of the model exceed the size of the training dataset. Due to this overparameterization, the training loss may have infinitely many global minima and it is critical to understand the properties of the solutions found by first-order optimization schemes such as (stochastic) gradient descent starting from different initializations. In this paper we demonstrate that when the loss has certain properties over a minimally small neighborhood of the initial point, first order methods such as (stochastic) gradient descent have a few intriguing properties: (1) the iterates converge at a geometric rate to a global optima even when the loss is nonconvex, (2) among all global optima of the loss the iterates converge to one with a near minimal distance to the initial point, (3) the iterates take a near direct route from the initial point to this global optima. As part of our proof technique, we introduce a new potential function which captures the precise tradeoff between the loss function and the distance to the initial point as the iterations progress. For Stochastic Gradient Descent (SGD), we develop novel martingale techniques that guarantee SGD never leaves a small neighborhood of the initialization, even with rather large learning rates. We demonstrate the utility of our general theory for a variety of problem domains spanning low-rank matrix recovery to neural network training. Underlying our analysis are novel insights that may have implications for training and generalization of more sophisticated learning problems including those involving deep neural network architectures.