The Power of Interpolation: Understanding the Effectiveness of SGD in Modern Over-parametrized Learning

TL;DR

This paper explains why stochastic gradient descent (SGD) converges quickly in modern over-parametrized models, showing the role of data interpolation and mini-batch size regimes, with theoretical bounds and experimental validation.

Contribution

It provides a formal analysis of SGD convergence in over-parametrized regimes, identifying critical mini-batch sizes and regimes, with explicit formulas for quadratic loss and experimental support.

Findings

Exponential convergence bounds for mini-batch SGD in convex settings.

Identification of a critical mini-batch size separating two regimes.

O(n) acceleration over full gradient descent per unit of computation.

Abstract

In this paper we aim to formally explain the phenomenon of fast convergence of SGD observed in modern machine learning. The key observation is that most modern learning architectures are over-parametrized and are trained to interpolate the data by driving the empirical loss (classification and regression) close to zero. While it is still unclear why these interpolated solutions perform well on test data, we show that these regimes allow for fast convergence of SGD, comparable in number of iterations to full gradient descent. For convex loss functions we obtain an exponential convergence bound for {\it mini-batch} SGD parallel to that for full gradient descent. We show that there is a critical batch size $m^*$ such that: (a) SGD iteration with mini-batch size $m\leq m^*$ is nearly equivalent to $m$ iterations of mini-batch size $1$ (\emph{linear scaling regime}). (b) SGD iteration with mini-batch $m> m^*$ is nearly equivalent to a full gradient descent iteration (\emph{saturation regime}). Moreover, for the quadratic loss, we derive explicit expressions for the optimal mini-batch and step size and explicitly characterize the two regimes above. The critical mini-batch size can be viewed as the limit for effective mini-batch parallelization. It is also nearly independent of the data size, implying $O(n)$ acceleration over GD per unit of computation. We give experimental evidence on real data which closely follows our theoretical analyses. Finally, we show how our results fit in the recent developments in training deep neural networks and discuss connections to adaptive rates for SGD and variance reduction.