Uncertainty Propagation in Deep Neural Networks Using Extended Kalman Filtering

TL;DR

This paper introduces a method using Extended Kalman Filtering to efficiently propagate and quantify uncertainty in deep neural networks, offering comparable accuracy with reduced computational costs and the ability to include model error.

Contribution

The paper presents an EKF-based approach for uncertainty propagation in DNNs that is computationally efficient and incorporates model error naturally.

Findings

Comparable uncertainty estimates to existing methods

Lower computational overhead

Inclusion of model error in uncertainty quantification

Abstract

Extended Kalman Filtering (EKF) can be used to propagate and quantify input uncertainty through a Deep Neural Network (DNN) assuming mild hypotheses on the input distribution. This methodology yields results comparable to existing methods of uncertainty propagation for DNNs while lowering the computational overhead considerably. Additionally, EKF allows model error to be naturally incorporated into the output uncertainty.