Continuous Development and Safety Assurance Pipeline for ML-Based Systems in the Railway Domain
摘要
Automated Driving Systems (ADS) will operate in an open world. Since it is difficult to specify all possible situations in this open world a priori and the world will most likely change during the system’s lifecycle, it requires agile MLOps cycles, including testing & validation in the field. In this paper, we present how the safe MLOps process proposed in Zeller et al. [25] is realized in the safe.trAIn project using Git-centric methods. Thereby, appropriate tooling support is provided at the different stages of the safe MLOps process to enable continuous development and safety assurance of ML-based systems in the railway domain.