Concept Enforcement and Modularization as Methods for the ISO 26262 Safety Argumentation of Neural Networks

Faculty/Professorship: University of Bamberg  
Author(s): Schwalbe, Gesina ; Schels, Martin
Title of the compilation: Proceeding of the 10th European Congress on Embedded Real Time Software and Systems
Corporate Body: European Congress on Embedded Real Time Software and Systems (ERTS), 10, 2020, Toulouse
Publisher Information: Toulouse
Year of publication: 2020
Pages: 1-10
Language(s): English
Neural networks (NN) are prone to systematic faults which are hard to detect using the methods recommended by the ISO 26262 automotive functional safety standard. In this paper we propose a unified approach to two methods for NN safety argumentation: Assignment of human interpretable concepts to the internal representation of NNs to enable modularization and formal verification. Feasibility of the required concept embedding analysis is demonstrated in a minimal example and important aspects for generalization are investigated. The contribution of the methods is derived from a proposed generic argumentation structure for a NN model safety case.
GND Keywords: ISO/DIS 26262 ; Funktionssicherheit ; Maschinelles Lernen ; Neurales Netzwerk ; Künstliche Intelligenz
Keywords: concept enforcement, machine learning, neural networks, functional safety, ISO 26262, goal structuring notation, explainable AI
DDC Classification: 004 Computer science  
RVK Classification: ST 300   
Peer Reviewed: Ja
International Distribution: Ja
Open Access Journal: Ja
Type: Conferenceobject
Release Date: 2. June 2020