You are responsible for managing and monitoring a Vertex AI model that is deployed in...

In a production MLOps pipeline, the goal is to move from manual retraining to Continuous Training (CT) triggered by real-world performance.

Vertex AI Model Monitoring: This service is specifically designed to monitor models in production for Prediction Drift and Feature Skew . It compares incoming production data to the training baseline. By integrating this with Cloud Pub/Sub or Vertex Pipelines, you can automate the retraining trigger when model performance (like accuracy or precision, if ground truth is available) or data distributions drift beyond a set threshold.

Why other options are incorrect:

Option A: Explainable AI is for understanding why a model made a prediction; while shifts in attribution can indicate drift, it is a proxy, not a direct measure of performance deterioration.

Option B: User feedback is subjective, often delayed, and difficult to programmatically use as a reliable trigger for automated retraining.

Option D: A static dataset does not represent the changing nature of production data. If the world changes (drift), the model might still perform well on the old static data but fail in the real world.