Variance-reduced Clipping for Non-convex Optimization

TL;DR

This paper introduces a variance-reduced clipping method for non-convex optimization that improves theoretical complexity bounds and demonstrates competitive empirical performance in deep learning tasks.

Contribution

It develops a variance reduction technique combined with gradient clipping under relaxed smoothness assumptions, achieving order-optimal complexity bounds.

Findings

Improves stochastic gradient complexity to O(ε^{-3}) using SPIDER.

Achieves order-optimal complexity for finite-sum problems with O(√n ε^{-2} + n).

Empirically outperforms or matches existing variance-reduced methods in vision tasks.

Abstract

Gradient clipping is a standard training technique used in deep learning applications such as large-scale language modeling to mitigate exploding gradients. Recent experimental studies have demonstrated a fairly special behavior in the smoothness of the training objective along its trajectory when trained with gradient clipping. That is, the smoothness grows with the gradient norm. This is in clear contrast to the well-established assumption in folklore non-convex optimization, a.k.a. $L$--smoothness, where the smoothness is assumed to be bounded by a constant $L$ globally. The recently introduced $(L_0,L_1)$--smoothness is a more relaxed notion that captures such behavior in non-convex optimization. In particular, it has been shown that under this relaxed smoothness assumption, SGD with clipping requires $O(\epsilon^{-4})$ stochastic gradient computations to find an $\epsilon$--stationary solution. In this paper, we employ a variance reduction technique, namely SPIDER, and demonstrate that for a carefully designed learning rate, this complexity is improved to $O(\epsilon^{-3})$ which is order-optimal. Our designed learning rate comprises the clipping technique to mitigate the growing smoothness. Moreover, when the objective function is the average of $n$ components, we improve the existing $O(n\epsilon^{-2})$ bound on the stochastic gradient complexity to $O(\sqrt{n} \epsilon^{-2} + n)$, which is order-optimal as well. In addition to being theoretically optimal, SPIDER with our designed parameters demonstrates comparable empirical performance against variance-reduced methods such as SVRG and SARAH in several vision tasks.