Cianplan

Level 5: AI / LLM systems · L5.8 Generative AI on AWS: Bedrock in practice (AIP-C01)

Bias and drift for predictive models: Clarify, Model Monitor and what replaced them

Measure bias before and after training, monitor a deployed model for data, quality, bias and attribution drift the way the exam expects (SageMaker Clarify and Model Monitor), and rebuild the same monitoring with AWS's named replacements now that both are closed to new customers.

13 min · Pro

What you take away

  • Bias is measured twice: in the data before training (class imbalance, DPL) and in the predictions after (DPPL, disparate impact), and a person has to choose which fairness metric matters.

And 3 more inside the class, in Pro.

How the class is laid out

  1. The problem
  2. The host's seating rule
  3. Measuring bias: before and after training
  4. Monitoring a deployed model
  5. Closed to new customers since 2026-07-30
  6. Check your understanding