17th Symposium - The Working Prozess of the Internal Combustion Engine
Architecture and independence controller for deep learning in safety critical applications
Authors
Dr. Ulrich Bodenhausen, Ulrich Bodenhausen AI Coaching and Vector Consulting Services GmbH
Year
2019
Summary
The market potential of safety critical products using AI is very attractive and Deep Learning Neural Networks (NN) have proven strengths to provide important functionality. This paper describes some of the challenges in arguing safety of systems using Deep Learning NN, especially functional improvement in context of SOTIF (Safety of the Intended Functionality) or other approaches to provide the safety case. An architecture and independence controller is proposed which can be used beneficially to reduce residual risk of functional insufficiencies for Deep Learning NN based systems.
Members of the Austrian Society of Automotive Engineers have access to all lectures of the International Vienna Motor Symposia.