AI-300 Practice Question 45: Monitor Event Grid Retraining Trigger

Model monitoring supports threshold-based alerts, and the documentation says that by using events generated by Azure Machine Learning model monitoring runs you can set up event-driven applications or CI/CD workflows with Azure Event Grid. When a monitor detects drift, data quality problems or model performance degradation, you can track these events with Event Grid and take action programmatically. The documentation's own example matches: if the accuracy of a classification model in production dips below a certain threshold, you can use Event Grid to begin a retraining job that uses collected ground truth data (option B).

- A is wrong because polling is the manual, delayed approach that the team wants to avoid, and it needs a scheduler and code to keep running. - C is wrong because an email only notifies people. It does not start anything, so retraining still depends on someone reading it. - D is wrong because a nightly schedule retrains regardless of the model's condition, wasting compute when nothing has changed and reacting too slowly when accuracy drops right after a run.

Reference: Model monitoring in production - Azure Machine Learning