Model deployment

Objectives

1. Integrate the machine learnt component into the target system in such a manner that the system satisfies the allocated system safety requirements. The component should be integrated in the pipeline linking its inputs and outputs to other system components.
2. Demonstrate that the allocated system safety requirements are still satisfied during operation of the target system and environment.
3. Create an assurance argument to demonstrate that the ML model will continue to meet the ML safety requirements once integrated into the target system.

Inputs to the stage

• [A]: System safety requirements
• [B]: Environment description
• [C]: System description
• [V]: ML model
• [GG]: ML deployment argument pattern
• [EE]: Operational scenarios

Outputs of the stage

• [DD] : Erroneous behaviour log
• [FF] : Integration testing results
• [HH] : ML deployment argument

Description of the stage

As shown in the AMLAS model deployment assurance process diagram above, this stage consists of three activities that provide a basis for ML component deployment assurance. This process shall be followed not only for the initial deployment of the component but also for any subsequent deployment required to update the component within the system. The artefacts generated from this stage are used to instantiate the ML model deployment assurance argument pattern as part of Activity 17.

Additional guidance on this stage can be found at [9].