Google Professional Machine Learning Engineer Question 494
Select 4Google Cloud PlatformYou are working on a credit scoring model hosted on Google Cloud, and the stakeholders have requested the ability to explain individual predictions for regulatory compliance. Which of the following steps should you take to implement Explainable AI and meet these requirements?
- A
Use Vertex AI's built-in Explainable AI feature to generate feature attributions for your model.
- B
Enable SHAP (SHapley Additive exPlanations) in Vertex AI to understand the impact of input features on predictions.
- C
Use Explainable AI with models trained exclusively on TensorFlow, as it does not support other frameworks.
- D
Export your model to a format supported by AI Explanations, such as TensorFlow SavedModel or Scikit-learn, if not already compatible.
- E
Configure the deployed model in Vertex AI to return explanation metadata along with predictions.
Show answer and explanation
Correct answers: A, B, D, E
Explanation
Explainable AI in Google Cloud helps machine learning engineers and stakeholders understand and interpret model predictions. This involves generating feature attributions, using techniques like SHAP, and ensuring the model is in a compatible format. Vertex AI supports a variety of frameworks and formats, and explanation metadata must be configured for real-time predictions. These steps are essential for meeting regulatory compliance and providing transparency in model predictions.
- A. Correct.
Correct: Vertex AI's Explainable AI provides built-in tools to generate feature attributions, which are crucial for explaining individual predictions.
- B. Correct.
Correct: SHAP is a supported technique in Vertex AI for calculating feature attributions and understanding the contribution of each input feature to the model's predictions.
- C. Incorrect.
Incorrect: Explainable AI supports multiple frameworks, including TensorFlow, Scikit-learn, and XGBoost. It is not limited to TensorFlow, so this statement is incorrect.
- D. Correct.
Correct: If your model is not in a supported format, exporting it to a compatible format (e.g., TensorFlow SavedModel or Scikit-learn) ensures Explainable AI can process it.
- E. Correct.
Correct: To provide explanations in real-time, the deployed model in Vertex AI must be configured to return explanation metadata along with predictions.