Google Professional Machine Learning Engineer Question 459
Select 3Google Cloud PlatformYou are tasked with designing a machine learning system on Google Cloud to predict customer purchasing behavior. To ensure the system is secure and protects against unintentional exploitation of sensitive customer data, which steps should you take?
- A
Implement differential privacy techniques to anonymize sensitive data during both training and inference.
- B
Ensure that the training dataset is balanced and representative of all customer demographics.
- C
Use Google Cloud's Data Loss Prevention (DLP) API to identify and redact personally identifiable information (PII) in the dataset.
- D
Apply robust access control policies to datasets and models using Identity and Access Management (IAM).
- E
Deactivate logging for all machine learning models to avoid storing sensitive data.
Show answer and explanation
Correct answers: A, C, D
Explanation
To protect against unintentional exploitation of sensitive data in machine learning systems, differential privacy, redacting sensitive data using tools like the DLP API, and applying robust access controls are critical steps. These measures ensure that sensitive information is not exposed or misused during data processing, training, or inference. Balancing the dataset and deactivating logging are unrelated to securing sensitive data.
- A. Correct.
Differential privacy techniques help protect against unintentional exploitation of sensitive data by ensuring individual data points cannot be identified, making it a critical step for building a secure AI system.
- B. Incorrect.
While ensuring a balanced dataset is important for fairness in predictions, it does not directly address the security concern of protecting sensitive data from unintentional exploitation.
- C. Correct.
The Data Loss Prevention (DLP) API is specifically designed to identify and protect sensitive data such as PII, which is a key step in building secure AI systems.
- D. Correct.
Using Identity and Access Management (IAM) to enforce strict access control ensures that only authorized individuals can access sensitive datasets and models, reducing the risk of data exploitation.
- E. Incorrect.
Deactivating logging for all models is not recommended, as logs are valuable for debugging and monitoring. Instead, sensitive data should be protected by anonymization or redaction in logs.