Oversee Model Governance PMI-CPMAI Practice Questions (Page 5)

Questions 21–25

1. 21foundation · easy

   When a data scientist trains a new iteration of a model, what should be recorded in the versioning system to ensure the iteration is fully reproducible?

   • AThe deployment environment and the model's API endpoint
   • BThe name of the data scientist who trained the model
   • CThe model's unique version identifier and the date it was trained
   • DThe training dataset, code, hyperparameters, and evaluation metrics
   Reveal Solution Discussion Select an answer first
2. 22application · medium

   A logistics company retrains its delivery-time prediction model every night. The retraining job sometimes fails because the data pipeline is still running, and the model is deployed with stale weights. The team wants to ensure the deployed model always uses the freshest data without manual intervention. What should they implement?

   • AA retraining schedule that starts the training job at a fixed time, independent of the data pipeline's completion status.
   • BA retraining trigger that fires only after the data pipeline signals completion, followed by automated validation and deployment.
   • CA retraining schedule that runs the job twice per night to increase the chance of success.
   • DA manual process where an engineer checks the data pipeline each morning and kicks off retraining if the data looks complete.
   Reveal Solution Discussion Select an answer first
4. 23application · medium

   A bank's credit-risk model was promoted from staging to production last quarter. The model owner now wants to retire an older version that has been shadow-scoring for two months. The compliance team requires that any retired model remain reproducible for regulatory audits. Which action best satisfies the requirement?

   • AArchive the old model's versioned artifacts, metadata, and code in a read-only registry location with a documented retirement date.
   • BRe-tag the old model as 'deprecated' in the registry but leave all its compute resources active.
   • CKeep the old model running in shadow mode indefinitely so it remains available for comparison.
   • DDelete the old model's artifacts and code from the registry, but keep the audit log of its deployment.
   Reveal Solution Discussion Select an answer first
5. 24foundation · easy

   What is the primary purpose of a formal change control process for AI models?

   • ATo automatically deploy any new model version to production
   • BTo ensure that all changes to models are reviewed, approved, and documented before implementation
   • CTo track the performance of models in production
   • DTo prevent any changes to models after they are deployed
   Reveal Solution Discussion Select an answer first
6. 25foundation · easy

   Which mechanism is essential for enforcing governance policies by providing a record of actions taken on a model?

   • AModel retraining
   • BData augmentation
   • CHyperparameter tuning
   • DAudit trails
   Reveal Solution Discussion Select an answer first