The argument pattern relating to this stage of the AMLAS process is shown in Figure 12 below. The key elements of the argument pattern are described below.
The top claim in this argument pattern is that the development of the learnt model ([V]) is sufficient. The sufficiency of the model learning process is argued through considering the appropriateness of the model development activities undertaken.
The argument strategy is to argue over the internal testing of the model performed during development as well as the development approach adopted. The appropriateness of the development activities is considered within the context of creating a model that both satisfies the ML safety requirements as well as meeting the additional constraints that are imposed on the model, such as performance and cost.
It must be demonstrated that the ML model that is selected satisfies the ML safety requirements. This is shown by using the internal test data ([O]) generated from Activity 7. The model must be shown to satisfy the ML safety requirements when this test data is applied. The internal testing claim is supported through evidence from the internal test results ([X]).
A justification must be provided that the results obtained from the internal testing are sufficient to indicate that the ML safety requirements are satisfied. This justification is provided in J4.1.
This claim considers the approach that has been adopted in developing the model. This claim is supported by claims regarding the type of model selected, the model parameters that are used, and the process that is applied.
It must be demonstrated that the type of model that is created in Activity 5 is appropriate for the given set of ML safety requirements and the other model constraints. The evidence for the type of model selected is captured in the model development log ([U]), which is used as evidence to support this claim.
It must be demonstrated that the parameters of the selected model have been appropriately tuned in Activity 5. The parameters must be shown to be appropriate for the given set of ML safety requirements. The rationale for how the model parameters are determined should be captured in the model development log ([U]), which is used as evidence to support this claim.
It must be demonstrated that the process is appropriate. As discussed in Activity 5, this will be a highly iterative process involving a number of decisions on each iteration, and the development of multiple models. The process will also involve decisions regarding the model architecture. The rationale for the process decisions should be included in the model development log ([U]) along with a justification for the appropriateness of the development tool chain used.