AWS Certified Machine Learning – SpecialtyMachine Learning Implementation and OperationsHard
A healthcare provider uses a machine learning model to assist in patient diagnosis. Due to strict regulatory compliance, every inference made by the model must be logged, and the logs must be immutable and verifiable for audit purposes over several years. The provider uses Amazon SageMaker for model deployment. Which AWS service integration with SageMaker would best meet these requirements?
- AAmazon CloudWatch Logs for real-time monitoring and log storage.
- BAmazon Kinesis Data Firehose to stream logs to Amazon S3 with integrity checks and long-term storage configurations.
- CAWS CloudTrail to log API calls related to endpoint invocation.
- DAmazon S3 for storing inference request and response payloads.
Show answer & explanationAnswer & explanation
Correct answer: B. Amazon Kinesis Data Firehose to stream logs to Amazon S3 with integrity checks and long-term storage configurations.
Amazon Kinesis Data Firehose can stream inference logs (request/response) to Amazon S3, where S3's immutability features (like S3 Object Lock) and versioning can be used for verifiable, long-term storage, meeting strict audit requirements. Firehose also provides data transformation and aggregation capabilities.
Why the other options are wrong
- A. CloudWatch Logs stores logs but does not inherently provide the immutability and verifiable integrity features required for strict audit compliance over several years.
- C. AWS CloudTrail logs API calls made to AWS services, including SageMaker, but it does not capture the actual inference request and response payloads, which are critical for model auditability.
- D. While S3 can store data, direct storage of raw inference payloads to S3 from SageMaker doesn't automatically provide the streaming, aggregation, and integrity features needed for a robust audit trail, unless coupled with other services.
ML Inference Auditing
The process of recording and verifying every input, output, and decision made by a machine learning model, ensuring compliance, transparency, and accountability.
- Requires immutable storage of inference logs.
- Logs must include request and response payloads.
- Essential for regulatory compliance and debugging.
Memory trick: Audit trails for ML decisions demand verifiable, immutable records.