A Simple Baseline for Bayesian Uncertainty in Deep Learning

TL;DR

SWAG offers a simple, scalable method for uncertainty estimation in deep learning by fitting a Gaussian to SGD iterates, improving calibration and out-of-sample detection across various tasks.

Contribution

The paper introduces SWAG, a novel approach that combines SWA with a Gaussian approximation for Bayesian uncertainty in deep neural networks.

Findings

SWAG approximates the true posterior distribution of neural network weights.

SWAG outperforms popular methods like MC dropout and SGLD in calibration and out-of-sample detection.

The method is scalable and effective across diverse deep learning tasks.

Abstract

We propose SWA-Gaussian (SWAG), a simple, scalable, and general purpose approach for uncertainty representation and calibration in deep learning. Stochastic Weight Averaging (SWA), which computes the first moment of stochastic gradient descent (SGD) iterates with a modified learning rate schedule, has recently been shown to improve generalization in deep learning. With SWAG, we fit a Gaussian using the SWA solution as the first moment and a low rank plus diagonal covariance also derived from the SGD iterates, forming an approximate posterior distribution over neural network weights; we then sample from this Gaussian distribution to perform Bayesian model averaging. We empirically find that SWAG approximates the shape of the true posterior, in accordance with results describing the stationary distribution of SGD iterates. Moreover, we demonstrate that SWAG performs well on a wide variety of tasks, including out of sample detection, calibration, and transfer learning, in comparison to many popular alternatives including MC dropout, KFAC Laplace, SGLD, and temperature scaling.