Google Professional Machine Learning Engineer Question 326
Select 3Google Cloud PlatformYou are working on a machine learning project that uses customer data to predict purchase behavior. The dataset includes sensitive information such as customer IDs and purchase histories. You decide to use Vertex AI Feature Store to manage features for this project. Which of the following actions would you take to ensure secure and efficient usage of Vertex AI Feature Store?
- A
Enable encryption for feature data stored in Vertex AI Feature Store using Customer-Managed Encryption Keys (CMEK).
- B
Use a unique Feature Store for each machine learning model to avoid sharing sensitive data across projects.
- C
Implement access control policies using IAM roles to restrict access to the Feature Store.
- D
Store raw sensitive data, such as customer IDs, directly in the Feature Store to simplify the process.
- E
Leverage online and offline serving capabilities to retrieve feature values based on the prediction use case.
Show answer and explanation
Correct answers: A, C, E
Explanation
Vertex AI Feature Store is designed to securely and efficiently manage features for machine learning projects. Enabling encryption with CMEK ensures data security, while IAM roles provide fine-grained access control. Leveraging online and offline serving capabilities supports a variety of prediction use cases. However, raw sensitive data should not be stored directly in the Feature Store, and creating a unique Feature Store for each model is unnecessary with proper feature-sharing practices.
- A. Correct.
Correct: Enabling encryption using CMEK ensures sensitive data is securely stored, meeting compliance and security requirements.
- B. Incorrect.
Incorrect: Creating a unique Feature Store for each model might lead to inefficiencies and unnecessary duplication of features. Features can be shared across models securely with proper access control.
- C. Correct.
Correct: Using IAM roles to implement access control ensures that only authorized users or services can access the Feature Store, enhancing security.
- D. Incorrect.
Incorrect: Storing raw sensitive data, such as customer IDs, directly in the Feature Store is not recommended. Sensitive data should be transformed (e.g., tokenized or hashed) to maintain privacy.
- E. Correct.
Correct: Leveraging online and offline serving capabilities allows efficient retrieval of feature values for real-time and batch prediction use cases.