Machine learning algorithms, especially Deep Neural Networks (DNNs), are becoming more widespread in the automotive industry particularly in computer vision applications such as perception due to their powerful performance. However, the question how to ensure safety of DNNs is still an open question. This presentation aims to:
• Analyze the development steps of DNNs i.e. Define, Specify, Develop and Evaluate, and Deploy and Monitor, from safety perspective.
• Discuss which safety measures need to be performed and which safety artifacts need to be generated in each development step.
• Discuss about strategies as redundancy and ensemble concepts at architectural level in order to meet the safety requirements of the overall system.