Google Professional Machine Learning Engineer Question 461
Select 4Google Cloud PlatformYou are designing a machine learning model for an e-commerce platform that predicts user preferences based on their past behavior. To ensure the system is secure and protects against unintentional exploitation of sensitive user data, which of the following approaches should you implement?
- A
Implement differential privacy techniques to ensure user data cannot be reverse-engineered from model outputs.
- B
Use feature selection to exclude sensitive attributes like race, gender, or financial information from the training data.
- C
Store all training data in a publicly accessible bucket to allow for scalability and ease of access.
- D
Conduct regular adversarial testing to identify and mitigate vulnerabilities in the system.
- E
Configure IAM policies to restrict access to the training datasets and model artifacts.
Show answer and explanation
Correct answers: A, B, D, E
Explanation
Building secure AI systems requires a combination of techniques to protect sensitive data, prevent unintentional exploitation, and mitigate vulnerabilities. Differential privacy, sensitive feature exclusion, adversarial testing, and proper IAM configuration are essential practices. Storing data in a publicly accessible location is a critical security violation and should be avoided.
- A. Correct.
Differential privacy techniques add noise to the data or model outputs, making it difficult for attackers to infer sensitive information from individual data points. This is a key strategy in building secure AI systems.
- B. Correct.
Excluding sensitive attributes from training data reduces the risk of the model unintentionally learning biases or making predictions that exploit sensitive data. This is a best practice for responsible AI development.
- C. Incorrect.
Storing training data in a publicly accessible bucket is a severe security risk, as it exposes sensitive user information to unauthorized access and potential exploitation.
- D. Correct.
Adversarial testing helps identify vulnerabilities in the system, such as susceptibility to adversarial attacks or data leakage, allowing you to address them proactively.
- E. Correct.
Configuring IAM policies ensures that only authorized users and applications have access to sensitive data and model artifacts, reducing the risk of unintentional or malicious misuse.