Towards Dependability Metrics for Neural Networks

Artificial neural networks (NN) are instrumental in realizing highly-automated driving functionality. An overarching challenge is to identify best safety engineering practices for NN and other learning-enabled components. In particular, there is an urgent need for an adequate set of metrics for measuring all- important NN dependability attributes. We address this challenge by proposing a number of NN-specific and efficiently computable metrics for measuring NN dependability attributes including robustness, interpretability, completeness, and correctness.

I. Introduction

Artificial neural networks (NN) are instrumental in realizing a number of important features in safety-relevant applications such as highly-automated driving. In particular, vision-based perception, the prediction of drivers' intention, and even end-to-end autonomous control are usually based on NN technology. State-of-the-practice safety engineering processes (cmp. ISO 26262) require that safety-relevant components, including NN-enabled ones, demonstrably satisfy their respective safety goals.