AIF-C01 Question 180
Select 3A company is deploying a machine learning model to predict loan eligibility for its customers. The compliance team is concerned about potential bias in the model and its impact on different population subgroups. Which AWS tools or techniques should the company use to detect and monitor bias in the model and ensure its trustworthiness?
- A
Amazon SageMaker Clarify to analyze bias in the data and model predictions
- B
Amazon SageMaker Model Monitor to track data drift and bias in real-time
- C
Amazon Augmented AI (A2I) to manually review predictions flagged as biased by machine learning models
- D
Amazon Rekognition to detect bias in image datasets
- E
Human audits to evaluate the fairness of model predictions
Show answer and explanation
Correct answers: A, B, E
Explanation
To detect and monitor bias in machine learning models, a combination of automated tools like Amazon SageMaker Clarify and SageMaker Model Monitor, along with human audits, is most effective. SageMaker Clarify provides pre-training and post-training bias metrics, while Model Monitor ensures ongoing fairness by detecting changes in data and predictions. Human audits complement these tools by providing qualitative assessments of fairness.
- A. Correct.
Amazon SageMaker Clarify helps analyze bias in datasets and model predictions by providing metrics to evaluate fairness, making it highly effective for addressing bias concerns.
- B. Correct.
Amazon SageMaker Model Monitor allows ongoing monitoring of deployed models for issues like data drift and bias, ensuring trustworthiness and fairness over time.
- C. Incorrect.
Amazon Augmented AI (A2I) is designed for human review of ML predictions but is not a tool specifically designed to analyze or monitor bias directly.
- D. Incorrect.
Amazon Rekognition is a computer vision service and does not provide functionality for detecting or addressing bias in datasets or models.
- E. Correct.
Human audits can be used to evaluate the fairness of model predictions and provide additional insights into potential bias from a qualitative perspective.