AIF-C01 Guidelines for Responsible AI Practice Question
Which TWO actions can help mitigate bias in a face recognition model trained on AWS? (Select two.)
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Ensure the training dataset is balanced across demographics
Options A and B are correct. Mitigating bias in a face recognition model requires a balanced training dataset (A) and regular evaluation of model performance across demographic subgroups (B). Option C (deploying in multiple regions) affects latency or availability, not bias. Option D (larger neural network) does not address data imbalance. Option E (Amazon Rekognition's content moderation) is for detecting inappropriate content, not for bias mitigation.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✓
Ensure the training dataset is balanced across demographics
Why this is correct
Balancing the training dataset across demographics directly addresses the stem's bias-mitigation constraint by preventing the model from overfitting to a majority group. Underrepresented subgroups otherwise yield skewed feature weights, so equalising sample counts per demographic reduces disparate error rates before training begins.
- ✓
Regularly evaluate model performance across subgroups
Why this is correct
Subgroup evaluation exposes disparate error rates that aggregate accuracy hides, satisfying the stem's bias-mitigation requirement. Measuring precision, recall and false-positive rates per demographic reveals which groups the model under-serves, enabling targeted retraining or threshold adjustment rather than blind redeployment.
- ✗
Deploy the model in multiple regions
Why it's wrong here
Regional deployment changes where inference runs, not the training data distribution that produces demographic bias. It is tempting because multi-region architectures address latency and resilience, and would be correct when the requirement is availability or data residency rather than fairness across face attributes.
- ✗
Use a larger neural network
Why it's wrong here
Adding layers or parameters increases model capacity without altering the imbalanced training data that causes bias, and can worsen minority-class performance. It is tempting because larger networks often raise overall accuracy, and would be correct when the constraint is underfitting on a balanced, representative dataset.
- ✗
Use Amazon Rekognition's content moderation
Why it's wrong here
Content moderation flags inappropriate imagery; it neither measures nor corrects demographic performance disparities in face matching. It is tempting because Rekognition offers it as a built-in safeguard, and it would be correct when filtering uploaded images for unsafe or explicit content rather than addressing training-data bias.
Go deeper
Related to this question
About these practice questions
This AIF-C01 question is part of Courseiva's 862-question bank — original exam-style content with full explanations and wrong-answer analysis, never real exam questions or exam dumps. Learn why practice questions differ from exam dumps →
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This AIF-C01 practice question is part of Courseiva's free Amazon Web Services certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the AIF-C01 exam.