MLPER-18: Include human-in-the-loop monitoring

Use human-in-the-loop monitoring to monitor model performance efficiently. When automating decision processes, the human labeling of model results is a reliable quality test for model inferences.

Compare human labels with model inferences to estimate model performance degradation. Perform mitigation as model re-training.

Implementation plan

• Use Amazon Augmented AI to get human review - Learn how to design a quality assurance system for model inferences. Establish a team of subject matter experts to audit model inference in production. Use Amazon Augmented AI (Amazon A2I) to get human review of low-confidence predictions or random prediction samples. Amazon A2I uses resources in IAM, SageMaker AI, and Amazon S3 to create and run your human review workflows.

Documents

• Using Amazon Augmented AI for Human Review

Blogs

• Human-in-the-loop review of model explanations with Amazon SageMaker AI Clarify and Amazon A2I

• Verifying and adjusting your data labels to create higher quality training datasets with Amazon SageMaker AI Ground Truth

Videos

• Easily Implement Human in the Loop into Your Machine Learning Predictions with Amazon A2I