Comparing the quality of neural network uncertainty estimates for classification problems

TL;DR

This paper evaluates and compares various uncertainty quantification methods for deep learning classifiers, highlighting the inconsistency among methods and emphasizing the need for principled quality assessment metrics.

Contribution

It introduces a framework for evaluating UQ methods in deep learning, comparing multiple approaches using statistical metrics on real and simulated data.

Findings

MCMC Bayesian neural networks perform best overall.

Bootstrapped neural networks are a close second in quality.

Different UQ methods can produce markedly different uncertainty estimates.

Abstract

Traditional deep learning (DL) models are powerful classifiers, but many approaches do not provide uncertainties for their estimates. Uncertainty quantification (UQ) methods for DL models have received increased attention in the literature due to their usefulness in decision making, particularly for high-consequence decisions. However, there has been little research done on how to evaluate the quality of such methods. We use statistical methods of frequentist interval coverage and interval width to evaluate the quality of credible intervals, and expected calibration error to evaluate classification predicted confidence. These metrics are evaluated on Bayesian neural networks (BNN) fit using Markov Chain Monte Carlo (MCMC) and variational inference (VI), bootstrapped neural networks (NN), Deep Ensembles (DE), and Monte Carlo (MC) dropout. We apply these different UQ for DL methods to a hyperspectral image target detection problem and show the inconsistency of the different methods' results and the necessity of a UQ quality metric. To reconcile these differences and choose a UQ method that appropriately quantifies the uncertainty, we create a simulated data set with fully parameterized probability distribution for a two-class classification problem. The gold standard MCMC performs the best overall, and the bootstrapped NN is a close second, requiring the same computational expense as DE. Through this comparison, we demonstrate that, for a given data set, different models can produce uncertainty estimates of markedly different quality. This in turn points to a great need for principled assessment methods of UQ quality in DL applications.