Make Me a BNN: A Simple Strategy for Estimating Bayesian Uncertainty from Pre-trained Models
This addresses the problem of uncertainty estimation for real-world computer vision applications, offering a scalable solution that is incremental by building on pre-trained models.
The paper tackles the challenge of reliable uncertainty quantification in deep neural networks (DNNs) by introducing the Adaptable Bayesian Neural Network (ABNN), a strategy that transforms pre-trained DNNs into Bayesian Neural Networks (BNNs) with minimal overhead, achieving state-of-the-art performance in image classification and semantic segmentation tasks.
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.