Out of Distribution Data Detection Using Dropout Bayesian Neural Networks

TL;DR

This paper investigates how dropout-based Bayesian neural networks can be used to detect out-of-distribution data, proposing a new uncertainty measurement method that improves detection across multiple tasks.

Contribution

It introduces a novel approach to measuring embedding uncertainty in dropout BNNs, enhancing OOD detection performance over previous methods.

Findings

Embedding uncertainty measurement improves OOD detection.

The method is effective across image, language, and malware detection tasks.

Theoretical justification supports the proposed approach.

Abstract

We explore the utility of information contained within a dropout based Bayesian neural network (BNN) for the task of detecting out of distribution (OOD) data. We first show how previous attempts to leverage the randomized embeddings induced by the intermediate layers of a dropout BNN can fail due to the distance metric used. We introduce an alternative approach to measuring embedding uncertainty, justify its use theoretically, and demonstrate how incorporating embedding uncertainty improves OOD data identification across three tasks: image classification, language classification, and malware detection.