Manage AI/ML Transparency (e.g., Data Selection, Algorithm Selection) PMI-CPMAI Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   A logistics company is training a model to predict delivery delays. The raw data contains timestamps with timezone inconsistencies, missing GPS coordinates, and duplicate entries. The data team cleans the data by standardizing timezones to UTC, imputing missing coordinates using the last known location, and removing duplicates. For an external audit, what MUST the preprocessing report document?

   • AThe names of the engineers who performed the data cleaning
   • BThe specific imputation method used for missing GPS coordinates
   • CThe number of records removed as duplicates and the imputation method for missing coordinates
   • DA step-by-step description of the timezone standardization, imputation method, and duplicate removal criteria
   Reveal Solution Discussion Select an answer first
2. 7foundation · easy

   When applying interpretability techniques to a real model, what is the goal of generating explanations for individual predictions?

   • ATo improve the model's overall accuracy
   • BTo replace the need for model validation
   • CTo understand why the model made a specific decision for a given input
   • DTo automatically correct biased predictions
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A cybersecurity firm uses a deep learning model to detect network intrusions. The security analysts need to understand why a specific network event was flagged as an intrusion so they can respond appropriately. The model is a black box. Which interpretability approach is MOST appropriate for this task?

   • AUse LIME to generate a local explanation for the specific network event
   • BUse the model's training accuracy to assess the reliability of the alert
   • CUse a simpler model (e.g., logistic regression) to replace the deep learning model
   • DUse a global feature importance plot to understand which features matter most across all events
   Reveal Solution Discussion Select an answer first
5. 9foundation · easy

   Why is it important to maintain audit trails for algorithmic decisions?

   • ATo reduce the need for model monitoring
   • BTo satisfy the requirement for model retraining
   • CTo enable accountability and review of decisions by logging the decision-making process
   • DTo improve the model's accuracy over time
   Reveal Solution Discussion Select an answer first
6. 10application · medium

   A government agency uses an AI system to screen benefit applications. The system is updated quarterly with new model versions. A citizen appeals a denial, and the agency must reconstruct exactly which model version made the decision, what inputs were used, and what the output was. What is the MINIMUM audit trail requirement to support this appeal?

   • ALog the training data used for each quarterly model update
   • BLog only the final decision (approved or denied) for each application
   • CLog the names of the data scientists who approved each model version
   • DLog the model version, the application inputs, the model output, and the timestamp for every decision
   Reveal Solution Discussion Select an answer first