AIF-C01 Question 182
Select 3A machine learning model trained for loan approval decisions has shown signs of potential bias against certain demographic groups. Which of the following tools or techniques can be used to detect and monitor bias in this model and ensure its trustworthiness?
- A
Amazon SageMaker Clarify
- B
Amazon Augmented AI (Amazon A2I)
- C
AWS Identity and Access Management (IAM)
- D
Subgroup analysis
- E
SageMaker Model Monitor
Show answer and explanation
Correct answers: A, D, E
Explanation
Detecting and monitoring bias in machine learning models is essential for ensuring fairness and trustworthiness. Amazon SageMaker Clarify specifically provides tools for bias detection and explainability. Subgroup analysis helps identify performance variations across demographic groups, which may indicate bias. SageMaker Model Monitor supports ongoing monitoring of deployed models to detect issues, including potential bias. While Amazon A2I and IAM are useful AWS tools, they do not directly address bias detection or monitoring.
- A. Correct.
Amazon SageMaker Clarify can be used to detect biases during data preparation, model training, and model inference by analyzing feature attributions and measuring bias metrics.
- B. Incorrect.
Amazon Augmented AI (Amazon A2I) is primarily used for adding human review workflows to ML predictions, but it is not specifically designed for detecting or monitoring bias.
- C. Incorrect.
AWS Identity and Access Management (IAM) is a security service for managing access to AWS resources and does not provide functionality for detecting or monitoring bias in machine learning models.
- D. Correct.
Subgroup analysis is a technique used to evaluate how well a model performs across different demographic groups, helping to detect bias and ensure fairness in model predictions.
- E. Correct.
SageMaker Model Monitor is used to continuously monitor deployed models for data drift, prediction quality issues, and other anomalies, which can help detect and address bias over time.