Abstract

This topic has been submitted as a request for the development of a DIN SPEC according to the PAS procedure. DIN SPECs are developed within DIN-SPEC-consortiums set up on a temporary basis. In the attached business plan you will find detailed information on the planned project as well as concrete time limits for commenting on the business plan (four-week commentary period) and for registering for the kick-off. This document will be developed and approved by the authors named in the business plan. This document defines methods and metrics that allow to measure the degradations in performance and health of AI-based systems and to detect their operational anomalies, including guidance regarding the following topics:

• Applicability of the method or metric for a given AI component of a system.
  • Suitability of the method or metric given a type of expected degradation.
    • Exemplary implementations of the methods and metrics for some select use cases.
      • Factual background information on statistical measures and their characteristics Cyber security-related monitoring is excluded from the scope of this specification and left for future work. This document is intended for "AI developers", "AI testers" and "AI auditors"

      .

Begin

2024-10-15

Planned document number

DIN SPEC 91527

Project number

62022049