An image classifier has higher error rates for one condition.
Investigate and improve it before release.
Choose two appropriate actions.
Analyze data, labels, and model causes, then measure group-level error rates again after mitigation.
Detailed explanation
Hiding the condition does not improve the user impact.
Choose a mitigation based on the cause and measure its effect.
High-impact or uncertain decisions may need human oversight.
Disparate errors become visible and can be tracked over time.
Check official documentation
Compare the following points with the official documentation.
AWS公式AIF-C01ガイドのDomain 4.1でバイアス検出・軽減・継続評価を確認する。
Official references
What to confirm
バイアスを測り、原因に対処し、改善後に再測定する流れを説明できる。
Key points
- Cause analysis
- Mitigation
- Re-evaluation
Notes
- Environment: AWS公式AIF-C01試験ガイドとAWS公式ドキュメントの確認
- Compare the definition with the AIF-C01 objective and the service documentation before applying it.
- Keep the concept separate from a particular answer choice or implementation step.
Foundation review
Mitigation loop
Detection, cause analysis, intervention, and remeasurement form a continuous improvement cycle.
Unintended effects
A mitigation can improve one group or metric while harming another, so multiple measures matter.