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.

• ✗

  The physical Azure data centre locations where model training takes place

  Why it's wrong here

  Azure regions refer to the physical clusters of data centers that host compute, storage, and networking services; they determine data residency, high availability, and latency. An Azure ML environment, by contrast, is not a location but a declarative specification of the software stack—Python version, pip/conda packages, and a base Docker image—used to launch a run. Training runs execute on hardware in some region, but that same environment specification can be reused identically across regions, so region choice has no bearing on dependency reproducibility.

• ✓

  Versioned software configurations (Python packages, dependencies) ensuring reproducible ML runs

  Why this is correct

  In Azure ML, an Environment is an immutable, versioned resource that captures the exact Python interpreter, required pip and conda packages, and Docker base image for a run. By pinning these dependencies under a name and version, every run can be reproduced byte-for-byte on any compute target, regardless of who launches it or when. Registered environments also support lineage tracking, so a model can be traced back to the precise software stack used to train it.

• ✗

  Development, staging, and production deployment targets for Azure ML models

  Why it's wrong here

  Development, staging, and production are deployment lifecycle stages in an MLOps pipeline, each representing a different model version or endpoint target for serving predictions. Azure ML Environments are a separate concept: they are versioned software recipes (for example, a conda dependencies YAML plus a Docker base image) that define how a training or scoring script executes. While you might use the same environment definition when deploying to multiple stages, the environment itself is not a deployment target or a slot in a CI/CD pipeline.

• ✗

  The security boundaries that isolate different ML projects in the same Azure subscription

  Why it's wrong here

  Security isolation among ML projects in a workspace is enforced through Azure RBAC assignments that limit who can read, write, or delete resources, and through network controls such as private endpoints and managed virtual networks. An Azure ML Environment, however, contains only software configuration—package lists and runtime settings—and carries no permissions or access policies. Merely using the same environment in multiple projects does not create any data exposure or violate isolation, because access checks are performed by the control plane, not by the environment.

Go deeper

Related to this question