Methodological Guidelines for Rulebased Monitoring

The models designed with AI technologies are not easily testable and certifiable due to models complexity and data quality requirements. This guide proposes methods to implement a monitor rule-based able to detect and explain deviation of the AI component deployed in production from the expected behavior and precursors of the occurrence of failure conditions based on a predefined set of safety properties.