MLOps

Model monitoring tracks shifts in input data distributions and prediction outputs over time to detect data and concept drift, ensuring the model remains accurate. While options like A and D represent standard infrastructure concerns, they do not address the statistical degradation of ML models that model monitoring is designed to catch.

Model explainability tools like Vertex Explainable AI help users understand model predictions by calculating feature attributions to show which inputs most influenced the output. In contrast, hardware utilization (C) and network paths (D) are infrastructure metrics managed by monitoring tools rather than ML explainability frameworks.

A feature store provides a centralized repository that ensures the exact same feature definitions and values are used during both training and real-time serving, preventing training-serving skew. A diagram for this topic isn't available yet — the worked example below walks the same reasoning step by step.

• !While options like A and D represent standard infrastructure concerns, they do not address the statistical degradation of ML models that model monitoring is designed to catch.
• !Option D is incorrect because comparing GPU types is a hardware benchmarking task rather than a method for evaluating live model prediction quality.

Answer each from memory — you'll see instantly whether you're right and why.

Run a focused 10-question mini-mock on MLOps and see it stick.