Objective 3.5 Address Business Risks, Ethical Concerns, and Related Concepts in Training and Tuning AIP-210 Practice Questions (Page 3)

Questions 11–15

1. 11foundation · easy

   Which technique adds controlled noise to training data or model outputs to protect individual privacy?

   • AFeature scaling
   • BDifferential privacy
   • CRegularization
   • DData augmentation
   Reveal Solution Discussion Select an answer first
2. 12foundation · easy

   Which regulation specifically addresses the protection of personal data of individuals in the European Union?

   • ASOX (Sarbanes-Oxley Act)
   • BCCPA (California Consumer Privacy Act)
   • CHIPAA (Health Insurance Portability and Accountability Act)
   • DGDPR (General Data Protection Regulation)
   Reveal Solution Discussion Select an answer first
4. 13application · medium

   A hiring platform uses a model to rank job applicants. The model was trained on resumes from the past five years. The company discovers that the model ranks female applicants lower than male applicants with similar qualifications. They want to reduce this bias while maintaining the model's ability to predict job performance. Which technique should they use?

   • AAdversarial debiasing, which trains the model to predict the target while preventing a discriminator from predicting the protected attribute.
   • BUsing a more complex model to capture subtle patterns.
   • COversampling the minority class in the training data.
   • DRemoving all features related to gender, such as name and hobbies.
   Reveal Solution Discussion Select an answer first
5. 14application · medium

   A lending company uses a complex gradient boosting model to approve loans. A customer was denied a loan and asked for an explanation. The company must provide a reason that is understandable to the customer. Which approach should they use?

   • AUse LIME to generate a local explanation for the specific denial decision.
   • BShow the customer the model's training data to demonstrate fairness.
   • CReport the global feature importance from the model's built-in feature importance scores.
   • DTell the customer that the decision is based on a proprietary algorithm and cannot be disclosed.
   Reveal Solution Discussion Select an answer first
6. 15expert · medium

   A healthcare provider is required by regulators to explain individual predictions to patients. The model is a deep neural network that achieves high accuracy. The data science team is considering using SHAP or LIME. The team also needs to document the explanations for audits. Which approach should they take?

   • AUse SHAP because it provides consistent and theoretically grounded explanations, and store the SHAP values for each prediction.
   • BUse LIME because it is faster to compute and easier to implement.
   • CUse global feature importance from the model's gradients.
   • DUse a simpler model, such as logistic regression, to make explanations easier.
   Reveal Solution Discussion Select an answer first