Painless step size adaptation for SGD

TL;DR

This paper introduces the LIGHT function for adaptive step size in SGD, improving neural network convergence and generalization without overparameterization or stability guarantees.

Contribution

It proposes a novel LIGHT function that explicitly enhances both convergence and generalization in neural networks, simplifying step size adaptation.

Findings

Improves convergence and generalization without stability guarantees

Enables more reliable and explainable network architectures

Reduces need for overparameterization

Abstract

Convergence and generalization are two crucial aspects of performance in neural networks. When analyzed separately, these properties may lead to contradictory results. Optimizing a convergence rate yields fast training, but does not guarantee the best generalization error. To avoid the conflict, recent studies suggest adopting a moderately large step size for optimizers, but the added value on the performance remains unclear. We propose the LIGHT function with the four configurations which regulate explicitly an improvement in convergence and generalization on testing. This contribution allows to: 1) improve both convergence and generalization of neural networks with no need to guarantee their stability; 2) build more reliable and explainable network architectures with no need for overparameterization. We refer to it as "painless" step size adaptation.