AI-900 Practice Question: Describe fundamental principles of machine learning on Azure

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  Watching the training loss curve during model training to detect overfitting

  Why it's wrong here

  Watching the training loss curve during model training is a training-time diagnostic used to detect underfitting, overfitting, or convergence issues, and to decide when to stop training. It happens before deployment, using historical data and labels, not on live production data. Model monitoring, by contrast, operates after deployment on real-world inference requests, tracking data drift and performance degradation that cannot be observed from a training loss curve.

• ✓

  Tracking deployed model performance and data drift over time to detect degradation

  Why this is correct

  Tracking deployed model performance and data drift over time is the essence of model monitoring: it continuously compares live inference data against the training dataset to detect feature-distribution drift and evaluates prediction quality using metrics like accuracy or AUC when ground truth is available. When drift triggers a threshold, it alerts data scientists to a silent degradation that could warrant retraining. This is a post-deployment, production-focused practice that Azure Machine Learning supports natively.

• ✗

  A dashboard showing the compute costs of running model inference in production

  Why it's wrong here

  A dashboard showing compute costs for inference is a financial/operational metric, typically handled by Cost Management and Azure Monitor, not by model monitoring. It tracks resource consumption (e.g., cost per request) rather than the quality of predictions or shifts in the underlying data distribution. Model monitoring specifically measures ML-related signals like data drift and model accuracy degradation, so this option describes infrastructure cost monitoring, not model monitoring.

• ✗

  Monitoring the uptime and latency of the model serving endpoint

  Why it's wrong here

  Endpoint uptime and latency are infrastructure SLA metrics that reflect the serving platform's health and responsiveness, often captured via Application Insights and load balancer diagnostics. These do not indicate whether the model's predictions remain correct or whether the input data has drifted from the training distribution. Model monitoring is concerned with the ML-specific quality of the deployed model, not the mechanics of the HTTP endpoint serving it.

Go deeper

Related to this question