Lookahead Optimizer: k steps forward, 1 step back

TL;DR

Lookahead is a new optimizer that enhances existing methods like SGD and Adam by iteratively updating two sets of weights, improving stability and performance with minimal additional cost.

Contribution

The paper introduces Lookahead, a novel optimization algorithm that improves training stability and performance of existing optimizers through a simple, orthogonal approach.

Findings

Lookahead improves training stability and reduces variance.

It enhances the performance of SGD and Adam on multiple benchmarks.

Minimal additional computational and memory overhead.

Abstract

The vast majority of successful deep neural networks are trained using variants of stochastic gradient descent (SGD) algorithms. Recent attempts to improve SGD can be broadly categorized into two approaches: (1) adaptive learning rate schemes, such as AdaGrad and Adam, and (2) accelerated schemes, such as heavy-ball and Nesterov momentum. In this paper, we propose a new optimization algorithm, Lookahead, that is orthogonal to these previous approaches and iteratively updates two sets of weights. Intuitively, the algorithm chooses a search direction by looking ahead at the sequence of fast weights generated by another optimizer. We show that Lookahead improves the learning stability and lowers the variance of its inner optimizer with negligible computation and memory cost. We empirically demonstrate Lookahead can significantly improve the performance of SGD and Adam, even with their default hyperparameter settings on ImageNet, CIFAR-10/100, neural machine translation, and Penn Treebank.