Data governance and lineage: AIF-C01 study notes and practice questions
What to know about Data governance and lineage for the AIF-C01 exam, from the Security, Compliance, and Governance for AI Solutions domain (14% of the exam guide), then 2 free questions to test it. Part of the AIF-C01 study guide.
During an audit, a regulator asks a car insurer to show exactly which dataset version, hyperparameters and processing steps produced the pricing model now in production, and to reproduce that model.
The data team retrains weekly from one shared dataset, and the record must be kept for seven years.
Which capability BEST provides this?
Choose one answer.
AA. SageMaker ML Lineage Tracking, which records the datasets, training job parameters and pipeline steps behind each model.
BB. AWS CloudTrail logs, which record who called the training APIs and when each call was made.
CC. AWS Config history of the training instances, which shows how their configuration settings changed over time.
DD. S3 Versioning on the training bucket, which keeps every earlier version of each object in the bucket.
B. Wrong: CloudTrail records API activity for auditing. It shows who called an action and when, but not which dataset and hyperparameters produced a model.
C. Wrong: Config tracks resource configurations, such as instance settings. It does not link a model to the datasets and parameters used to train it.
D. Wrong: Versioning keeps old copies of the data, which helps. It does not record the training job, its parameters or the steps that connected the data to the model.
Explanation
SageMaker ML Lineage Tracking records the steps and artifacts of an ML workflow, from data preparation to deployment, so a team can trace and reproduce a model. Audit logs, configuration history and object versions each answer a different question. This question tests choosing lineage over general audit tools.
A European bank uses foundation models on Amazon Bedrock.
Its data residency policy requires that prompts and responses are processed only within the EU. Traffic peaks sometimes exceed what one Region can serve, so the bank needs cross-Region inference.
Which approach meets both requirements?
Choose one answer.
AA. Use an EU geographic cross-Region inference profile, which routes requests only among Regions within the EU.
BB. Use a global cross-Region inference profile, which can spread requests across Regions beyond any single geography.
CC. Use one in-Region on-demand model only, and accept that requests above the Region's capacity are throttled at peak times.
DD. Use a US geographic cross-Region inference profile, since it offers more capacity than the EU profile.
B. Wrong: A global profile is not limited to one geography, so it can send requests to Regions outside the EU. That breaks the residency rule.
C. Wrong: A single Region keeps data in the EU but cannot absorb peak traffic beyond its capacity. The bank's throughput need is not met.
D. Wrong: A US profile routes requests only among US Regions, which moves EU prompts outside the EU. It fails the residency rule.
Explanation
Cross-Region inference raises throughput by routing requests across Regions. Geographic profiles restrict that routing to one geography, such as the EU or the US, which is how residency rules are met. This question tests combining a throughput need with a residency limit.
Want more on Data governance and lineage? The 5 full-length practice sets have 13 more Data governance and lineage questions, 325 in all, none of them in the free exam. One payment of $14.99, no subscription. See the practice sets →