Microsoft AI-300 (Operationalizing Machine Learning and Generative AI Solutions) Practice Test 23

Question 220

An AI risk review evaluates whether a predictive model treats important populations consistently. The immediate design goal is to compare an approved mitigation with the baseline under both equity and predictive performance metrics. Which approach addresses that goal?

  1. Fairness mitigation experiment
  2. Protected-feature leakage check
  3. Fairness documentation
  4. Data representation audit

Explanation

Correct Answer

A:

Mitigation can alter tradeoffs between group metrics and other performance dimensions. It should be evaluated empirically rather than assumed harmless.

Incorrect Answers

B:

Proxy analysis investigates indirect pathways to disparate outcomes; dropping the protected attribute does not guarantee fair treatment; this approach leaves the team unable to compare an approved mitigation with the baseline under both equity and predictive performance metrics.

C:

Documented results make assumptions and residual risks reviewable; a one-number model score cannot describe all fairness properties; the needed capability is to compare an approved mitigation with the baseline under both equity and predictive performance metrics.

D:

Representation checks examine whether available training data covers intended users; post-training threshold changes alone cannot repair missing evidence for a population; the requirement, by contrast, is to compare an approved mitigation with the baseline under both equity and predictive performance metrics.

Question 221

An AI risk review evaluates whether a predictive model treats important populations consistently. The team is required to record evaluated groups, metrics, known limitations, and stakeholder decisions in the model release evidence. Which option is the best fit?

  1. Fairness mitigation experiment
  2. Data representation audit
  3. Performance parity assessment
  4. Fairness documentation

Explanation

Correct Answer

D:

Documented results make assumptions and residual risks reviewable. A one-number model score cannot describe all fairness properties.

Incorrect Answers

A:

Mitigation can alter tradeoffs between group metrics and other performance dimensions; it should be evaluated empirically rather than assumed harmless; the requirement, by contrast, is to record evaluated groups, metrics, known limitations, and stakeholder decisions in the model release evidence.

B:

Representation checks examine whether available training data covers intended users; post-training threshold changes alone cannot repair missing evidence for a population; the team would still need to record evaluated groups, metrics, known limitations, and stakeholder decisions in the model release evidence.

C:

Segment-level parity analysis reveals uneven error distributions; the aggregate accuracy metric may conceal that disparity; the specified outcome instead requires the team to record evaluated groups, metrics, known limitations, and stakeholder decisions in the model release evidence.

Question 222

An AI risk review evaluates whether a predictive model treats important populations consistently. An operator is tasked to look for proxy variables that could reproduce sensitive information even when an explicit column was removed. Which mechanism provides that capability?

  1. Group fairness analysis
  2. Data representation audit
  3. Protected-feature leakage check
  4. Performance parity assessment

Explanation

Correct Answer

C:

Proxy analysis investigates indirect pathways to disparate outcomes. Dropping the protected attribute does not guarantee fair treatment.

Incorrect Answers

A:

Fairness analysis tests distributional performance differences across groups; it requires suitable labels, context, and careful interpretation of chosen fairness criteria; the implementation under review needs to look for proxy variables that could reproduce sensitive information even when an explicit column was removed.

B:

Representation checks examine whether available training data covers intended users; post-training threshold changes alone cannot repair missing evidence for a population; the requested task remains to look for proxy variables that could reproduce sensitive information even when an explicit column was removed.

D:

Segment-level parity analysis reveals uneven error distributions; the aggregate accuracy metric may conceal that disparity; the actual design criterion is to look for proxy variables that could reproduce sensitive information even when an explicit column was removed.

Question 223

A regulated lender needs to understand and review the drivers of model decisions. The immediate design goal is to identify which features most influence predictions across the evaluated population. Which approach addresses that goal?

  1. Global feature importance
  2. Counterfactual analysis
  3. Error analysis dashboard
  4. Local feature explanation

Explanation

Correct Answer

A:

Global importance summarizes broad model behavior. It should not be mistaken for a definitive reason behind one individual prediction.

Incorrect Answers

B:

Counterfactual explanations examine changes associated with a different predicted outcome; they require constraints to avoid suggesting impossible or inappropriate actions; the outstanding need is to identify which features most influence predictions across the evaluated population.

C:

Error analysis segments mistakes to reveal clusters that global evaluation can obscure; it guides investigation but is not itself an automatic repair; this approach leaves the team unable to identify which features most influence predictions across the evaluated population.

D:

Local explanations attribute a particular prediction to input features under a chosen explanation method; they differ from aggregate importance rankings; the needed capability is to identify which features most influence predictions across the evaluated population.

Question 224

A regulated lender needs to understand and review the drivers of model decisions. The team is required to explain the factors that contributed to one particular applicant's model score. Which option is the best fit?

  1. Model interpretability tradeoff
  2. Local feature explanation
  3. Global feature importance
  4. Error analysis dashboard
  5. Responsible AI report

Explanation

Correct Answer

B:

Local explanations attribute a particular prediction to input features under a chosen explanation method. They differ from aggregate importance rankings.

Incorrect Answers

A:

Interpretable models can simplify review when prediction quality is otherwise adequate; complex models may need additional explanation tools and governance controls; the needed capability is to explain the factors that contributed to one particular applicant's model score.

C:

Global importance summarizes broad model behavior; it should not be mistaken for a definitive reason behind one individual prediction; the team would still need to explain the factors that contributed to one particular applicant's model score.

D:

Error analysis segments mistakes to reveal clusters that global evaluation can obscure; it guides investigation but is not itself an automatic repair; the specified outcome instead requires the team to explain the factors that contributed to one particular applicant's model score.

E:

A responsible AI report documents risks and model behavior beyond an accuracy chart; reviewers still need to assess whether evidence meets domain obligations; the implementation under review needs to explain the factors that contributed to one particular applicant's model score.

Question 225

A regulated lender needs to understand and review the drivers of model decisions. An operator is tasked to explore what actionable input changes could alter a particular model decision within permitted constraints. Which mechanism provides that capability?

  1. Local feature explanation
  2. Error analysis dashboard
  3. Model interpretability tradeoff
  4. Counterfactual analysis

Explanation

Correct Answer

D:

Counterfactual explanations examine changes associated with a different predicted outcome. They require constraints to avoid suggesting impossible or inappropriate actions.

Incorrect Answers

A:

Local explanations attribute a particular prediction to input features under a chosen explanation method; they differ from aggregate importance rankings; the specified outcome instead requires the team to explore what actionable input changes could alter a particular model decision within permitted constraints.

B:

Error analysis segments mistakes to reveal clusters that global evaluation can obscure; it guides investigation but is not itself an automatic repair; the implementation under review needs to explore what actionable input changes could alter a particular model decision within permitted constraints.

C:

Interpretable models can simplify review when prediction quality is otherwise adequate; complex models may need additional explanation tools and governance controls; the requested task remains to explore what actionable input changes could alter a particular model decision within permitted constraints.

Question 226

A regulated lender needs to understand and review the drivers of model decisions. The immediate design goal is to choose a transparent modeling approach when regulators demand readily understandable decision logic. Which approach addresses that goal?

  1. Global feature importance
  2. Local feature explanation
  3. Model interpretability tradeoff
  4. Counterfactual analysis

Explanation

Correct Answer

C:

Interpretable models can simplify review when prediction quality is otherwise adequate. Complex models may need additional explanation tools and governance controls.

Incorrect Answers

A:

Global importance summarizes broad model behavior; it should not be mistaken for a definitive reason behind one individual prediction; this alternative does not achieve the requirement to choose a transparent modeling approach when regulators demand readily understandable decision logic.

B:

Local explanations attribute a particular prediction to input features under a chosen explanation method; they differ from aggregate importance rankings; the actual design criterion is to choose a transparent modeling approach when regulators demand readily understandable decision logic.

D:

Counterfactual explanations examine changes associated with a different predicted outcome; they require constraints to avoid suggesting impossible or inappropriate actions; this approach leaves the team unable to choose a transparent modeling approach when regulators demand readily understandable decision logic.

Question 227

A regulated lender needs to understand and review the drivers of model decisions. The team is required to locate recurring false-negative patterns by cohort and feature ranges for targeted remediation. Which option is the best fit?

  1. Error analysis dashboard
  2. Responsible AI report
  3. Counterfactual analysis
  4. Model interpretability tradeoff

Explanation

Correct Answer

A:

Error analysis segments mistakes to reveal clusters that global evaluation can obscure. It guides investigation but is not itself an automatic repair.

Incorrect Answers

B:

A responsible AI report documents risks and model behavior beyond an accuracy chart; reviewers still need to assess whether evidence meets domain obligations; the needed capability is to locate recurring false-negative patterns by cohort and feature ranges for targeted remediation.

C:

Counterfactual explanations examine changes associated with a different predicted outcome; they require constraints to avoid suggesting impossible or inappropriate actions; the requirement, by contrast, is to locate recurring false-negative patterns by cohort and feature ranges for targeted remediation.

D:

Interpretable models can simplify review when prediction quality is otherwise adequate; complex models may need additional explanation tools and governance controls; the team would still need to locate recurring false-negative patterns by cohort and feature ranges for targeted remediation.

Question 228

A regulated lender needs to understand and review the drivers of model decisions. An operator is tasked to preserve evaluation evidence of explanations, fairness findings, and limitations for approval review. Which mechanism provides that capability?

  1. Responsible AI report
  2. Global feature importance
  3. Local feature explanation
  4. Model interpretability tradeoff

Explanation

Correct Answer

A:

A responsible AI report documents risks and model behavior beyond an accuracy chart. Reviewers still need to assess whether evidence meets domain obligations.

Incorrect Answers

B:

Global importance summarizes broad model behavior; it should not be mistaken for a definitive reason behind one individual prediction; the specified outcome instead requires the team to preserve evaluation evidence of explanations, fairness findings, and limitations for approval review.

C:

Local explanations attribute a particular prediction to input features under a chosen explanation method; they differ from aggregate importance rankings; the implementation under review needs to preserve evaluation evidence of explanations, fairness findings, and limitations for approval review.

D:

Interpretable models can simplify review when prediction quality is otherwise adequate; complex models may need additional explanation tools and governance controls; the requested task remains to preserve evaluation evidence of explanations, fairness findings, and limitations for approval review.

Question 229

An ML service must return predictions synchronously to application requests. The immediate design goal is to expose a trained model through an HTTPS-style real-time prediction interface with managed serving infrastructure. Which approach addresses that goal?

  1. Managed online endpoint
  2. Endpoint request timeout
  3. Online deployment instance count
  4. Deployment under endpoint

Explanation

Correct Answer

A:

A managed online endpoint handles synchronous requests using deployed inference code and compute. It differs from batch processing that writes results for large offline inputs.

Incorrect Answers

B:

Timeout settings bound how long requests can run before failing; extending them may mask slow scoring code and is not a substitute for profiling; this alternative does not achieve the requirement to expose a trained model through an HTTPS-style real-time prediction interface with managed serving infrastructure.

C:

Replica count controls deployed serving capacity within available quotas; it does not alter model accuracy or evaluation-set construction; the actual design criterion is to expose a trained model through an HTTPS-style real-time prediction interface with managed serving infrastructure.

D:

An online deployment packages inference assets behind an endpoint; multiple deployments can support safe rollout without changing the client-facing endpoint name; the outstanding need is to expose a trained model through an HTTPS-style real-time prediction interface with managed serving infrastructure.

Question 230

An ML service must return predictions synchronously to application requests. The team is required to run a particular model, scoring code, and runtime configuration as a named implementation of one endpoint. Which option is the best fit?

  1. Endpoint request timeout
  2. Endpoint traffic allocation
  3. Managed online endpoint
  4. Deployment under endpoint

Explanation

Correct Answer

D:

An online deployment packages inference assets behind an endpoint. Multiple deployments can support safe rollout without changing the client-facing endpoint name.

Incorrect Answers

A:

Timeout settings bound how long requests can run before failing; extending them may mask slow scoring code and is not a substitute for profiling; the outstanding need is to run a particular model, scoring code, and runtime configuration as a named implementation of one endpoint.

B:

Traffic weights support canary and blue/green testing among deployments; merely registering another model version does not change live request routing; this approach leaves the team unable to run a particular model, scoring code, and runtime configuration as a named implementation of one endpoint.

C:

A managed online endpoint handles synchronous requests using deployed inference code and compute; it differs from batch processing that writes results for large offline inputs; the needed capability is to run a particular model, scoring code, and runtime configuration as a named implementation of one endpoint.

Operationalizing Machine Learning and Generative AI Solutions (AI-300) Practice Tests