Google Professional Machine Learning Engineer Question 471
Select 3Google Cloud PlatformYou are tasked with developing a machine learning model for a financial institution to predict loan approvals. During testing, you discover that the model disproportionately denies loans to applicants from certain demographic groups. Which of the following actions should you take to align with Google’s Responsible AI practices?
- A
Analyze the training dataset for potential bias and ensure balanced representation from all demographic groups.
- B
Implement continuous monitoring of the model’s predictions to detect and mitigate bias over time.
- C
Deploy the model immediately since it meets accuracy benchmarks, and document the bias for future improvement.
- D
Use techniques like reweighting or adversarial debiasing to mitigate the identified bias before deploying the model.
- E
Disable demographic-based features entirely to ensure fairness in predictions.
Show answer and explanation
Correct answers: A, B, D
Explanation
To align with Google's Responsible AI practices, it is essential to proactively address bias in machine learning models. This includes analyzing the training data for bias, applying mitigation techniques, and continuously monitoring the model's behavior for fairness. Ignoring known bias or taking overly simplistic measures, such as removing demographic features without analysis, can result in suboptimal outcomes and violations of Responsible AI principles.
- A. Correct.
Analyzing the training dataset for potential bias is a fundamental step in identifying and addressing issues of underrepresentation or overrepresentation in data, which can lead to biased predictions.
- B. Correct.
Continuous monitoring is crucial to ensure that bias does not emerge or worsen in the model over time, aligning with Responsible AI practices.
- C. Incorrect.
Deploying the model immediately despite known bias contradicts Responsible AI principles, as it risks perpetuating harm and unfair treatment.
- D. Correct.
Using debiasing techniques such as reweighting or adversarial debiasing is a recommended method for mitigating bias in machine learning models before deployment.
- E. Incorrect.
Disabling demographic-based features entirely can lead to unintended consequences, such as proxy variables indirectly introducing bias. Fairness often requires thoughtful handling of such features rather than outright removal.