Google Professional Machine Learning Engineer Question 460
Select 3Google Cloud PlatformYou are building an AI system that processes sensitive healthcare data to train a machine learning model. To ensure the system is secure and protects against unintentional exploitation of this data, which steps should you take during the development process?
- A
Implement differential privacy techniques to anonymize training data.
- B
Ensure that users of the model have access to the raw training data for transparency.
- C
Use role-based access control (RBAC) to restrict access to sensitive data.
- D
Regularly audit your training data for biases and potential privacy risks.
- E
Store sensitive training data in plain text for easier debugging.
Show answer and explanation
Correct answers: A, C, D
Explanation
Building secure AI systems involves implementing privacy-preserving techniques, such as differential privacy, using robust access control mechanisms like RBAC, and proactively auditing data for biases and privacy risks. These measures help protect sensitive information and prevent unintentional exploitation of data. Avoid practices like exposing raw training data or storing data in plain text, as they compromise security and privacy.
- A. Correct.
Differential privacy techniques help anonymize sensitive data by adding noise to the data, ensuring that individual data points cannot be re-identified, making this a critical step for secure AI systems.
- B. Incorrect.
Providing raw training data to users violates privacy principles and increases the risk of unintentional exploitation of sensitive information. Transparency should be achieved through other means, such as model explainability, not by exposing the raw data.
- C. Correct.
Role-based access control (RBAC) is a best practice for securing sensitive data by ensuring that only authorized individuals can access it, reducing the risk of data breaches.
- D. Correct.
Auditing data for biases and privacy risks helps identify potential vulnerabilities and ensures compliance with data protection regulations, making it a necessary step in building secure AI systems.
- E. Incorrect.
Storing sensitive data in plain text is a significant security risk because it makes the data vulnerable to unauthorized access and breaches. Data should always be encrypted at rest and in transit.