Document the proxy-bias and overfitting findings: Which action best fits a training-and-testing
Issues discovered in training and testing, such as proxy bias and overfitting, must be documented and remediated in-process before the model advances.
The question
During training of a loan-approval model, the team observes that a protected-attribute proxy is driving predictions and that validation accuracy is far higher than training accuracy on one subgroup. Governance requires issues found during training and testing to be identified and treated before the model advances. Which action best fits a training-and-testing risk-management response?
Preparing for AIGP? Take the free 5-min readiness quiz →
- Defer the concern to post-market monitoring after release so production data can confirm whether the subgroup disparity is truly material.Plausible but wrong: post-market monitoring is a real control, yet a known training-time defect must be handled during training/testing, not deferred to production.
- Escalate only the overfitting signal to the model owner and treat the proxy-attribute correlation as an acceptable business feature for now.Almost right because escalation is appropriate, but ignoring the proxy-bias half of the finding leaves a known fairness risk unmanaged during testing.
- Document the proxy-bias and overfitting findings, run bias and interpretability testing, and remediate before promoting the model onward. ✓Correct because III.B.4 requires identifying issues surfaced in training/testing and managing them (including bias and overfitting) before the model progresses in the lifecycle.
- Freeze the project and route the entire issue to the vendor due-diligence team for contractual remediation under the license agreement.Plausible but wrong: vendor due diligence applies to acquired systems, not to defects in a model the team is itself training and testing.
The trap
Believing production monitoring can substitute for remediating a defect already known during training and testing. How to remember it
Issues discovered in training and testing, such as proxy bias and overfitting, must be documented and remediated in-process before the model advances.
How many of these would you get right?
One of 1581 AIGP questions on Certsqill. Take a free five-minute check and see your score per domain — not one number, but which section to open tonight.
Test your AIGP readiness — freeMore Understanding How to Govern AI Development questions
- Validation and test data held out from training: To produce this evidence, on which data should the decisive →
- A record of datasets: Which artifact most directly satisfies this need? →
- A model card summarizing intended use: Which readiness artifact directly addresses the standardized summary →
- All 426 Understanding How to Govern AI Development questions →
Part of the Certsqill AIGP question bank · Understanding How to Govern AI Development ·
Every answer, right and wrong, comes with its own explanation.
