A logistics company has a model predicting delivery delays. The model was trained on data from the past two years. Recently, the company expanded to a new geographic region, and the model's performance has degraded. The feature distributions for delivery distance and package weight have shifted significantly. What type of drift is the model experiencing, and what should the ML engineer do?

Feature attribution drift; investigate why the model's feature importance changed.

Training-serving skew; fix the data preprocessing pipeline.

Concept drift; retrain the model with data from the new region.

Data drift; retrain the model with data that includes the new region.

A company has deployed a Gemini-powered email-drafting assistant. The ML team is monitoring the outputs and has noticed that the model occasionally generates responses that are grammatically correct but contain subtle factual errors about the company's products. The team wants to implement a monitoring system that catches these errors in production. What is the most effective approach?

Monitor the model's perplexity on generated emails to detect low-confidence outputs.

Track the input prompt distribution for data drift to identify unusual requests.

Use a factual-consistency metric that compares generated claims against a knowledge base of product information.

Deploy a second, larger model to review all outputs before they are sent.

What is the purpose of prompt testing in generative AI solutions?

To measure the model's inference speed

To test the model's resistance to adversarial attacks

To compare the model against industry benchmarks

To evaluate how different prompts affect the quality and relevance of outputs

A healthcare company has fine-tuned a Gemini model to generate patient education materials. They need to evaluate the model's outputs for medical accuracy and safety before deployment. They have a small set of expert-reviewed reference materials. What is the most appropriate evaluation approach?

Deploy the model and monitor user feedback for any complaints.

Use the model's confidence scores to filter out low-confidence outputs.

Use a combination of automated metrics (e.g., factual consistency) and expert human review on a sample of outputs.

Rely solely on BLEU scores against the expert-reviewed reference materials.

Google CloudProfessional Machine Learning Engineer

Domain 6Objective 2

Monitoring, Testing, and Troubleshooting AI Solutions PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 1)

Part of the Monitoring AI solutions domain, which accounts for ~13% of the PROFESSIONAL-MACHINE-LEARNING-ENGINEER exam.

33questions here

7free pages

8concepts

~13%of the exam

Questions 1–5

1application · medium
A logistics company has a model predicting delivery delays. The model was trained on data from the past two years. Recently, the company expanded to a new geographic region, and the model's performance has degraded. The feature distributions for delivery distance and package weight have shifted significantly. What type of drift is the model experiencing, and what should the ML engineer do?
Select an answer first
2expert · hard
A company has deployed a Gemini-powered email-drafting assistant. The ML team is monitoring the outputs and has noticed that the model occasionally generates responses that are grammatically correct but contain subtle factual errors about the company's products. The team wants to implement a monitoring system that catches these errors in production. What is the most effective approach?
Select an answer first
3foundation · easy
What is the purpose of prompt testing in generative AI solutions?
Select an answer first
4expert · hard
A healthcare company has fine-tuned a Gemini model to generate patient education materials. They need to evaluate the model's outputs for medical accuracy and safety before deployment. They have a small set of expert-reviewed reference materials. What is the most appropriate evaluation approach?
Select an answer first
5foundation · easy
A data scientist notices that a model's predictions are less accurate in production than during training. They suspect the distribution of a feature differs between the training dataset and the live serving data. What is this phenomenon called?
Select an answer first

Finished these 5 questions?

Review the revealed explanations, or continue through the curriculum.

Next 5 questions →Back to PROFESSIONAL-MACHINE-LEARNING-ENGINEER

Free Basic Practice is a study aid with revealable answers — not a scored exam. Examers.io is independent and not affiliated with or endorsed by Google Cloud. “PROFESSIONAL-MACHINE-LEARNING-ENGINEER” is a trademark of its owner, used for identification only.

Monitoring, Testing, and Troubleshooting AI Solutions PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice Questions (Page 1)

Questions 1–5

What is the purpose of prompt testing in generative AI solutions?

A healthcare company has fine-tuned a Gemini model to generate patient education materials. They need to evaluate the model's outputs for medical accuracy and safety before deployment. They have a small set of expert-reviewed reference materials. What is the most appropriate evaluation approach?

A data scientist notices that a model's predictions are less accurate in production than during training. They suspect the distribution of a feature differs between the training dataset and the live serving data. What is this phenomenon called?

Keep studying — Monitoring AI solutions