Manage Accountability Documentation and Audit Trail PMI-CPMAI Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   An e-commerce company is developing a recommendation system. The team considered two algorithms: a collaborative filtering model and a deep learning model. They chose the deep learning model because it showed a 5% improvement in click-through rate in offline tests. However, they did not document why they rejected the collaborative filtering approach. Later, a new team member questions the choice. What documentation should have been created?

   • AA stakeholder approval form signed by the project sponsor.
   • BA decision record that includes the rationale for choosing deep learning, the alternatives considered, and the evaluation results.
   • CAn email to the team explaining the choice, without any formal record.
   • DA version control commit that only includes the final model code.
   Reveal Solution Discussion Select an answer first
2. 7foundation · easy

   Which element is essential to include in an AI model development decision record?

   • AThe list of all software licenses used in the project
   • BThe alternatives that were considered and why they were rejected
   • CThe personal opinions of the development team about the model's performance
   • DThe marketing strategy for promoting the model to customers
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A healthcare startup is developing a diagnostic AI model. During a retrospective, the data science lead realizes that the team changed the loss function mid-training to improve accuracy but did not record the change. The team also used a cleaned version of the training dataset without documenting the cleaning steps. The startup must now reproduce the model for an external audit. What should the team do first?

   • ASubmit the existing model for regulatory approval without mentioning the undocumented change.
   • BDelete the current model and dataset versions and start a new project with strict version control.
   • CRe-train the model from scratch using the current dataset and document all steps going forward.
   • DReconstruct the decision record by interviewing team members and correlating with version control logs and dataset snapshots.
   Reveal Solution Discussion Select an answer first
5. 9foundation · easy

   Which of the following is a typical go/no-go decision point in the AI lifecycle?

   • ABefore deploying the model to production
   • BAfter the model has been retired
   • CWhen writing the final report to regulators
   • DDuring the initial brainstorming session
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   A research institute is training a model on a sensitive dataset that was obtained from a partner organization under a data sharing agreement. The agreement requires that the data be used only for the specified research purpose and that any copies be tracked. The institute's data science team has downloaded the dataset to multiple local machines and shared it via a cloud drive. What is the primary risk in this scenario?

   • AThe cloud drive will automatically make the data public.
   • BThe data may become corrupted due to multiple downloads.
   • CThe model will be less accurate because the data is used on multiple machines.
   • DLoss of chain of custody over the data, making it impossible to demonstrate compliant usage and track all copies.
   Reveal Solution Discussion Select an answer first