SmoothOut: Smoothing Out Sharp Minima to Improve Generalization in Deep Learning

TL;DR

This paper introduces SmoothOut, a novel framework that smooths sharp minima in deep neural networks through noise injection and averaging, leading to improved generalization in training.

Contribution

The paper proposes SmoothOut and its variants, which effectively eliminate sharp minima and enhance generalization, differing from existing noise injection methods by applying de-noising and adaptive noise strategies.

Findings

SmoothOut effectively eliminates sharp minima.

SmoothOut and AdaSmoothOut improve generalization across various training settings.

The methods outperform state-of-the-art solutions in experiments.

Abstract

In Deep Learning, Stochastic Gradient Descent (SGD) is usually selected as a training method because of its efficiency; however, recently, a problem in SGD gains research interest: sharp minima in Deep Neural Networks (DNNs) have poor generalization; especially, large-batch SGD tends to converge to sharp minima. It becomes an open question whether escaping sharp minima can improve the generalization. To answer this question, we propose SmoothOut framework to smooth out sharp minima in DNNs and thereby improve generalization. In a nutshell, SmoothOut perturbs multiple copies of the DNN by noise injection and averages these copies. Injecting noises to SGD is widely used in the literature, but SmoothOut differs in lots of ways: (1) a de-noising process is applied before parameter updating; (2) noise strength is adapted to filter norm; (3) an alternative interpretation on the advantage of noise injection, from the perspective of sharpness and generalization; (4) usage of uniform noise instead of Gaussian noise. We prove that SmoothOut can eliminate sharp minima. Training multiple DNN copies is inefficient, we further propose an unbiased stochastic SmoothOut which only introduces the overhead of noise injecting and de-noising per batch. An adaptive variant of SmoothOut, AdaSmoothOut, is also proposed to improve generalization. In a variety of experiments, SmoothOut and AdaSmoothOut consistently improve generalization in both small-batch and large-batch training on the top of state-of-the-art solutions.