Make Me a BNN: A Simple Strategy for Estimating Bayesian Uncertainty from Pre-trained Models

TL;DR

This paper introduces ABNN, a simple and scalable post-hoc method to transform pre-trained deep neural networks into Bayesian neural networks, significantly improving uncertainty estimation with minimal additional training.

Contribution

The paper proposes ABNN, a novel post-hoc adaptation technique that enables existing DNNs to perform Bayesian uncertainty estimation efficiently without retraining from scratch.

Findings

ABNN achieves state-of-the-art uncertainty quantification performance.

The method requires minimal computational overhead.

It performs well across image classification and segmentation tasks.

Abstract

Deep Neural Networks (DNNs) are powerful tools for various computer vision tasks, yet they often struggle with reliable uncertainty quantification - a critical requirement for real-world applications. Bayesian Neural Networks (BNN) are equipped for uncertainty estimation but cannot scale to large DNNs that are highly unstable to train. To address this challenge, we introduce the Adaptable Bayesian Neural Network (ABNN), a simple and scalable strategy to seamlessly transform DNNs into BNNs in a post-hoc manner with minimal computational and training overheads. ABNN preserves the main predictive properties of DNNs while enhancing their uncertainty quantification abilities through simple BNN adaptation layers (attached to normalization layers) and a few fine-tuning steps on pre-trained models. We conduct extensive experiments across multiple datasets for image classification and semantic segmentation tasks, and our results demonstrate that ABNN achieves state-of-the-art performance without the computational budget typically associated with ensemble methods.