Model verification

Objectives

1. Demonstrate that the model will meet the ML safety requirements when exposed to inputs not present during the development of the model.
2. Create an assurance argument, based on the evidence generated by the first objective. The argument should clearly demonstrate the relationship between the verification evidence and the ML safety requirements. It should explicitly explain the tradeoffs, assumptions and uncertainties concerning the verification results and the process by which they were generated.

Inputs to the stage

• [H]: ML safety requirements
• [P]: Verification data
• [V]: ML model
• [BB]: ML verification argument pattern

Outputs of the stage

• [Z]: ML verification results
• [AA]: Verification log
• [CC]: ML verification argument

Description of the stage

As shown in the AMLAS model verification assurance process diagram above, this stage consists of two activities that are performed to provide assurance in the ML model verification process. The primary artefacts generated from this stage are ML model verification results, which are used to instantiate the ML verification argument pattern as part of Activity 14.

Additional guidance on this stage can be found at [10].