Google Professional Machine Learning Engineer Question 466
Select 2Google Cloud PlatformYou are deploying a machine learning model on Google Cloud to detect fraudulent transactions for a financial services company. During testing, you find that an attacker is able to manipulate the model's predictions by injecting adversarial examples into the input data. Which of the following strategies should you implement to mitigate this type of attack?
- A
Train the model using adversarial training with perturbed examples.
- B
Enable VPC Service Controls to restrict network access to the model.
- C
Implement input sanitization to detect and reject anomalous inputs.
- D
Use Explainable AI to analyze predictions and identify vulnerabilities.
- E
Deploy the model behind a Cloud Armor policy to block malicious traffic.
Show answer and explanation
Correct answers: A, C
Explanation
Adversarial examples are deliberately crafted inputs designed to manipulate a machine learning model's predictions. To mitigate such attacks, adversarial training enhances the model's robustness by exposing it to these examples during training. Input sanitization further strengthens the defense by detecting and blocking anomalous or adversarial inputs before they interact with the model. While other options like Explainable AI and VPC Service Controls are valuable in different contexts, they do not directly address the specific challenge of adversarial input manipulation.
- A. Correct.
Adversarial training involves augmenting the training data with adversarial examples, making the model more robust against such attacks. This is a key mitigation mechanism for adversarial input manipulation.
- B. Incorrect.
While VPC Service Controls are effective for securing data and service interactions, they do not directly address the issue of adversarial examples manipulating the model's predictions.
- C. Correct.
Input sanitization helps identify and reject anomalous or adversarial inputs before they reach the model, limiting the attacker's ability to exploit vulnerabilities.
- D. Incorrect.
Explainable AI is useful for interpreting model behavior and identifying general vulnerabilities but does not directly mitigate adversarial attacks.
- E. Incorrect.
Cloud Armor is designed for protecting web applications from network-based attacks, such as DDoS, and does not mitigate adversarial attacks targeting a machine learning model's inputs.