AI Model Training, Testing, and Validation AAIR Practice Questions (Page 7)

Questions 31–35

1. 31foundation · easy

   Why is comprehensive documentation of training and testing activities important for AI governance?

   • AIt eliminates the need for ongoing model monitoring
   • BIt supports auditability and provides a record of decisions and changes
   • CIt increases the model's accuracy
   • DIt guarantees the model will be free from bias
   Reveal Solution Discussion Select an answer first
2. 32foundation · easy

   What is the primary purpose of model validation?

   • ATo reduce the model's complexity
   • BTo replace the need for ongoing monitoring
   • CTo increase the model's accuracy on the training set
   • DTo confirm that the model meets business and risk requirements
   Reveal Solution Discussion Select an answer first
4. 33application · medium

   A pharmaceutical company is validating a model that predicts drug side effects. The regulatory body requires evidence that the model meets predefined safety thresholds before approval. Which validation activity is most directly required?

   • ACompare the model's performance to a previous version of the model
   • BDocument the model's architecture and training algorithm
   • CEnsure the model achieves high accuracy on the training set
   • DTest the model against the predefined safety thresholds on a held-out test set and document the results
   Reveal Solution Discussion Select an answer first
5. 34foundation · easy

   What is data leakage in the context of AI model training?

   • AThe loss of training data due to storage failures
   • BThe model being too complex for the available data
   • CThe model's predictions being too sensitive to small input changes
   • DInformation from the test set inadvertently influencing the training process
   Reveal Solution Discussion Select an answer first
6. 35expert · hard

   A hiring platform's resume-screening model shows 92% accuracy overall, but the accuracy for female candidates is 78% while for male candidates it is 96%. The legal team requires the model to demonstrate fair treatment across protected groups. Which metric set should the risk team use to evaluate and document fairness?

   • ACompare false-positive and false-negative rates across gender groups, along with equalized odds
   • BRetrain the model on a dataset that excludes gender as a feature
   • CReport only the overall accuracy of 92% to show the model is generally effective
   • DUse a larger test set with more female candidates to increase statistical power
   Reveal Solution Discussion Select an answer first