AML Model Deployment & MLOps Flashcards

What is data lineage in an MLOps context?

Answer: The documented history of data from its origin through all transformations to its final use in a model

Data lineage tracks where data came from, how it was transformed, and how it was used, enabling auditability and debugging of ML pipelines.

What is the primary function of a model monitoring dashboard in production?

Answer: To track production metrics like prediction drift, latency, and data quality in real time

Monitoring dashboards give operations teams real-time visibility into model health, enabling rapid detection of and response to performance degradation.

Which automated retraining trigger strategy is considered a best practice in production MLOps?

Answer: Scheduled or event-driven retraining triggered by performance degradation thresholds or new data availability

Automated retraining pipelines that trigger on both schedules and performance thresholds keep models current without requiring manual intervention.

What is training-serving skew in machine learning deployment?

Answer: Discrepancies between how features are computed during offline training versus online model serving

Training-serving skew occurs when feature computation logic differs between offline training and online serving, causing unexpected prediction errors in production.

What does model quantization achieve during ML model deployment?

Answer: It reduces model size and speeds up inference by using lower-precision numerical representations

Quantization converts 32-bit floating point weights to 8-bit integers, drastically reducing model size and inference latency with minimal accuracy loss.

What is the role of an ML model registry in the MLOps lifecycle?

Answer: To serve as a central catalog for tracking, versioning, and managing trained model artifacts through the deployment lifecycle

A model registry provides a governed repository where teams can register, version, stage, and promote models from development to production.