Calibration of Deep Probabilistic Models with Decoupled Bayesian Neural Networks

TL;DR

This paper introduces a decoupled Bayesian Neural Network approach to improve the calibration of deep neural network outputs, enhancing their reliability in critical decision-making scenarios.

Contribution

It proposes a novel decoupled Bayesian stage using BNNs to calibrate DNN probabilities, demonstrating improved calibration over existing methods.

Findings

Enhanced calibration accuracy across multiple benchmarks

Robustness and flexibility of the proposed method

Outperforms several state-of-the-art calibration techniques

Abstract

Deep Neural Networks (DNNs) have achieved state-of-the-art accuracy performance in many tasks. However, recent works have pointed out that the outputs provided by these models are not well-calibrated, seriously limiting their use in critical decision scenarios. In this work, we propose to use a decoupled Bayesian stage, implemented with a Bayesian Neural Network (BNN), to map the uncalibrated probabilities provided by a DNN to calibrated ones, consistently improving calibration. Our results evidence that incorporating uncertainty provides more reliable probabilistic models, a critical condition for achieving good calibration. We report a generous collection of experimental results using high-accuracy DNNs in standardized image classification benchmarks, showing the good performance, flexibility and robust behavior of our approach with respect to several state-of-the-art calibration methods. Code for reproducibility is provided.