Dissecting Adam: The Sign, Magnitude and Variance of Stochastic Gradients

TL;DR

This paper analyzes the ADAM optimizer by dissecting its use of gradient sign and variance, revealing insights into its behavior and proposing a new method based on variance adaptation to improve performance.

Contribution

It provides a detailed interpretation of ADAM's mechanisms, isolates the problematic sign aspect, and introduces a variance-based adaptation method to enhance optimization.

Findings

Sign aspect of ADAM can harm generalization.

Variance adaptation improves SGD performance.

Disentangling ADAM's components offers new insights.

Abstract

The ADAM optimizer is exceedingly popular in the deep learning community. Often it works very well, sometimes it doesn't. Why? We interpret ADAM as a combination of two aspects: for each weight, the update direction is determined by the sign of stochastic gradients, whereas the update magnitude is determined by an estimate of their relative variance. We disentangle these two aspects and analyze them in isolation, gaining insight into the mechanisms underlying ADAM. This analysis also extends recent results on adverse effects of ADAM on generalization, isolating the sign aspect as the problematic one. Transferring the variance adaptation to SGD gives rise to a novel method, completing the practitioner's toolbox for problems where ADAM fails.