Data management

Objectives

1. Develop data requirements that are sufficient to allow for the ML safety requirements to be encoded as features against which the data sets to be produced in this stage may be assessed.
2. Generate data sets in accordance with the data requirements for use in the development and verification stages, providing a rationale for those activities undertaken with respect to the ML safety requirements.
3. Analyse the data sets obtained by objective 2 to determine their sufficiency in meeting the data requirements.
4. Create an assurance argument, based on the evidence generated by meeting the first three objectives, that provides a clear justification of the ML data requirements. This should explicitly state the assumptions and tradeoffs made and any uncertainties concerning the data requirements and the processes by which they were developed and validated.

Inputs to the stage

• [H] : ML safety requirements
• [R] : ML data argument pattern

Outputs of the stage

• [L] : Data requirements
• [M] : Data requirements justification report
• [N] : Development data
• [O] : Internal test data
• [P] : Verification data
• [Q] : Data generation log
• [S] : ML data validation results
• [T] : ML data argument

Description of the stage

As shown in the AMLAS ML data requirements assurance process diagram above, this stage consists of four activities that are performed to provide assurance in the ML data. The artefacts generated from this stage are used to instantiate the ML data assurance argument pattern as part of Activity 9.

Additional guidance on this stage can be found at [8].