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.

• ✗

  Network pipelines for transferring data between Azure regions at high speed

  Why it's wrong here

  In Azure, high-speed cross-region data transfer is handled by network services such as ExpressRoute, Azure Virtual WAN, or Azure Data Box, which provide connectivity or physical shipping rather than an execution engine for ML code. An Azure Machine Learning pipeline, by contrast, is a directed graph of compute steps—script runs, data transformations, model training, and evaluation—that manages dependencies and intermediate outputs. Confusing these two is common because both use the word 'pipeline,' but one is network infrastructure and the other is orchestrated software workflow.

• ✓

  Reusable orchestrated workflows that automate and version-control the full ML training lifecycle

  Why this is correct

  An Azure Machine Learning pipeline chains discrete, reusable steps into an orchestrated workflow that automates the full training lifecycle—from data preparation and feature engineering through model training, validation, and registration. Each step is versioned, and pipeline definitions track source code, inputs, outputs, and compute targets, making runs reproducible and auditable. Caching of unchanged steps and scheduled execution enable efficient retraining, which is why pipelines are central to production MLOps.

• ✗

  CI/CD pipelines in Azure DevOps for deploying application code to production

  Why it's wrong here

  Azure DevOps pipelines (Azure Pipelines) are continuous integration and delivery workflows that build application code, run unit tests, and deploy artifacts to staging or production environments. They operate primarily in the application delivery domain, not the ML experimentation/training domain. Azure Machine Learning pipelines solve a different problem: orchestrating model training and evaluation steps and can, in fact, be invoked from an Azure DevOps release pipeline as part of a larger CI/CD process.

• ✗

  Data pipelines that ingest streaming data from IoT sensors into Azure storage

  Why it's wrong here

  Streaming ingestion from IoT sensors typically uses services such as Azure Event Hubs, Azure IoT Hub, and Azure Stream Analytics, which process high-velocity events and land them in storage before any training begins. An Azure Machine Learning pipeline, by contrast, expects data to already reside in a versioned dataset or datastore and then executes a sequence of training-related transformations. While both involve data movement, the streaming pipeline provides real-time ingestion while the ML pipeline provides offline reproducibility for model-building steps.

Go deeper

Related to this question