Bayesian Neural Network Versus Ex-Post Calibration For Prediction Uncertainty

TL;DR

This paper compares Bayesian neural networks and post-hoc calibration methods for predictive uncertainty, showing Bayesian approaches are competitive, especially in small data and tabular datasets.

Contribution

It provides a comprehensive empirical comparison between Bayesian neural networks and calibration methods across diverse datasets, highlighting the effectiveness of Bayesian models.

Findings

Bayesian neural networks perform competitively with calibrated neural networks.

Bayesian models are particularly effective in small data and tabular settings.

Calibration methods remain a viable alternative for uncertainty estimation.

Abstract

Probabilistic predictions from neural networks which account for predictive uncertainty during classification is crucial in many real-world and high-impact decision making settings. However, in practice most datasets are trained on non-probabilistic neural networks which by default do not capture this inherent uncertainty. This well-known problem has led to the development of post-hoc calibration procedures, such as Platt scaling (logistic), isotonic and beta calibration, which transforms the scores into well calibrated empirical probabilities. A plausible alternative to the calibration approach is to use Bayesian neural networks, which directly models a predictive distribution. Although they have been applied to images and text datasets, they have seen limited adoption in the tabular and small data regime. In this paper, we demonstrate that Bayesian neural networks yields competitive performance when compared to calibrated neural networks and conduct experiments across a wide array of datasets.