Google Professional Machine Learning Engineer Question 473
Select 3Google Cloud PlatformYou are building a machine learning model on Google Cloud to automate loan approval decisions for a financial institution. During the evaluation phase, you notice that the model's predictions are significantly less favorable for applicants from certain demographic groups. Which actions should you take to align with Google’s Responsible AI practices?
- A
Analyze the model's training data for potential sources of bias and rebalance the dataset if necessary.
- B
Use Explainable AI tools in Vertex AI to understand which features are influencing the model's predictions.
- C
Deploy the model as is and monitor it for fairness concerns after deployment.
- D
Incorporate counterfactual fairness techniques to test how changes in sensitive attributes affect predictions.
- E
Remove all demographic features from the training data to ensure fairness.
Show answer and explanation
Correct answers: A, B, D
Explanation
To align with Google’s Responsible AI practices, you must proactively identify and mitigate bias in your machine learning model. This includes analyzing the training data for bias, using tools like Explainable AI to understand model behavior, and testing fairness through counterfactual analysis. Deploying a model with known fairness issues or relying solely on feature removal without further analysis does not meet Responsible AI standards.
- A. Correct.
Correct: Analyzing training data for sources of bias is a key step in identifying and mitigating fairness issues, ensuring compliance with Responsible AI practices.
- B. Correct.
Correct: Explainable AI tools help uncover how features influence predictions, allowing you to identify potential biases in the model's decision-making process.
- C. Incorrect.
Incorrect: Deploying the model without addressing known fairness issues violates Responsible AI practices. Bias should be mitigated before deployment.
- D. Correct.
Correct: Counterfactual fairness techniques allow you to evaluate how sensitive attributes impact predictions, helping to ensure fairness in the model.
- E. Incorrect.
Incorrect: Simply removing demographic features does not guarantee fairness, as bias can still be encoded in other correlated features.