Reliable Industrial AI

Description

The project “Reliability of machine learning methods in an industrial environment” deals with issues relating to the safe use of current ML methods. New complex application scenarios such as autonomous production plants or decentralized power generation can only be implemented using learning systems, but their reliability and safety must be ensured at all times. Testing the reliability of such adaptive, cyber-physical systems is currently still a major challenge. This raises various issues relating to the dependability of such ML-based applications, such as the correct handling of uncertainty in statistical predictions, options for flexible, adaptive testing and runtime monitoring of the decisions made by the system. It is also necessary to translate classically defined safety requirements into a structure that the learning system can understand. The aim is always to fully utilize the flexibility of the systems that can be achieved using current AI methods, without having to compromise on ensuring reliability. The project is being carried out in cooperation with Siemens AG.