Prepare and test your skills
Prepare and test your skills
Free sample question, answered and explained below. The Generative AI Leader practice bank has 2,191 more like it. Try 10 free: instant score, no account.
An enterprise deploys a predictive model to a live production endpoint on Google Cloud to forecast user engagement. Over several months, consumer engagement habits gradually evolve, and the operations team is concerned that shifting input data may degrade prediction quality. The team wants to configure automated monitoring that compares the statistical distribution of incoming production features against a baseline of historical production data collected in the past.
Which monitoring feature should the team implement to identify this statistical deviation over time?
Infrastructure tracking focuses on standard operational telemetry such as request latency (p95/p99), CPU/GPU resource utilization, request throughput (QPS), and HTTP error response codes (e.g., 4xx and 5xx errors) via Cloud Monitoring.
The team needed to monitor endpoint availability, ensure prediction latency stayed within service level agreements (SLAs), or detect server-side container crashes.
Inference drift detection is a continuous model monitoring capability in Google Cloud designed to track whether incoming production feature distributions change significantly over time. Unlike static checks, drift detection establishes a statistical baseline using past production data and continuously evaluates recently served live inference requests against that baseline.

Inference drift detection is specifically designed to identify gradual or sudden changes in production distributions where the reference baseline is prior operational data, fulfilling all scenario requirements precisely.
Feature attribution monitoring evaluates shifts in the relative importance or contribution of individual input features to the model's final predictions, commonly quantified using SHAP (SHapley Additive exPlanations) values.
The team wanted to track whether a key predictive feature (such as customer account age) unexpectedly dropped in relative importance when generating inference outcomes.
Training-serving skew detection is a monitoring mechanism that detects discrepancies between the data distribution used during model training and the data distribution encountered when serving predictions in production.
The organization needed to verify whether production data immediately following deployment matched the original dataset used during the model training phase.
Answer them yourself and get your score the moment you finish. No account needed.
Full access includes
Keep the momentum going with these hand-picked practice scenarios
Last updated