Objective 3.5 Address Business Risks, Ethical Concerns, and Related Concepts in Training and Tuning AIP-210 Practice Questions (Page 1)

Questions 1–5

1. 1foundation · easy

   What is the purpose of maintaining thorough documentation of model training and tuning decisions?

   • ATo provide a record for audits and to support accountability for model outcomes.
   • BTo reduce the need for monitoring after deployment.
   • CTo increase the model's accuracy by recording hyperparameter values.
   • DTo automatically generate code for model deployment.
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A government agency is developing a model to allocate public resources. The agency must ensure that the model development process is transparent and that decisions can be reviewed. They want to establish clear accountability for model outcomes. Which practice should they implement?

   • AUse a black-box model and rely on external consultants to explain it.
   • BKeep the model's decision process secret to avoid public scrutiny.
   • CAllow any data scientist to modify the model without review to speed up development.
   • DAssign a model risk owner and create a model documentation template that includes intended use, limitations, and evaluation results.
   Reveal Solution Discussion Select an answer first
4. 3foundation · easy

   Which fairness metric measures whether a model's positive prediction rate is the same across different groups?

   • AF1 score
   • BCalibration
   • CDemographic parity
   • DEqualized odds
   Reveal Solution Discussion Select an answer first
5. 4application · medium

   A multinational corporation is training a model on customer data from the EU and California. The legal team requires that the model not retain personal data longer than necessary and that individuals can request deletion of their data. Which practice should the ML team implement?

   • AStore all training data indefinitely to allow future model retraining.
   • BUse only synthetic data generated from the original dataset to avoid legal issues.
   • CAnonymize all personal data before training so that deletion requests are unnecessary.
   • DImplement a data retention policy that deletes personal data after a defined period and supports erasure requests.
   Reveal Solution Discussion Select an answer first
6. 5application · medium

   A research institute is training a model on medical images that include patient faces. The images are sensitive and must be protected. The institute wants to share the trained model with external collaborators without exposing patient data. Which approach should they use?

   • AShare the raw images with collaborators under a confidentiality agreement.
   • BTrain the model using federated learning, where the model is trained across multiple sites without sharing raw images.
   • CUse a pre-trained model and fine-tune it on a small public dataset.
   • DCrop out the faces from the images before training.
   Reveal Solution Discussion Select an answer first