Efficient variational Bayesian neural network ensembles for outlier detection

TL;DR

This paper introduces an efficient method for outlier detection using ensembles of Bayesian neural networks with variational approximation, achieving competitive results with existing ensembling techniques.

Contribution

The paper proposes a novel variational Bayesian neural network ensemble method specifically designed for outlier detection, combining efficiency with competitive performance.

Findings

Outlier detection performance is comparable to other efficient ensembling methods.

The proposed method effectively approximates the true posterior using gradient descent sampling.

Ensembles improve outlier detection accuracy in Bayesian neural networks.

Abstract

In this work we perform outlier detection using ensembles of neural networks obtained by variational approximation of the posterior in a Bayesian neural network setting. The variational parameters are obtained by sampling from the true posterior by gradient descent. We show our outlier detection results are comparable to those obtained using other efficient ensembling methods.