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.

• ✗

  Training ML models through an online learning management system

  Why it's wrong here

  An LMS such as Moodle or Coursera is a platform for delivering courses, not for training ML models. In machine learning, "online" refers to a specific incremental training algorithm that updates model weights sequentially as each new data example or mini-batch arrives, often using stochastic gradient descent. The confusion stems from the dual meaning of "online" as internet-connected versus data-streaming; LMS-based learning is purely educational delivery and has no mechanism for parameter updates or loss optimization.

• ✓

  Continuously updating model weights on new data as it arrives rather than batch retraining

  Why this is correct

  This is the precise definition of online learning, also called incremental learning: instead of periodically retraining on the entire historical dataset, the model's weights are adjusted continuously with each new data point or small batch using an update rule like gradient descent. This approach adapts quickly to non-stationary data distributions and is ideal for real-time streams such as fraud detection, sensor telemetry, or clickstream feedback. However, a critical drawback is catastrophic forgetting, where the model may overwrite previously acquired patterns when the incoming data distribution shifts sharply.

• ✗

  Requiring an internet connection during model training for cloud compute access

  Why it's wrong here

  Internet connectivity is an infrastructure requirement for accessing remote compute, not a property of the learning algorithm. Online learning can run entirely on edge devices or on-premises servers without any network connection, while batch training can be executed in the cloud over the internet. The term "online" in ML has historical roots in the sequential, stream-oriented processing of data points one at a time, which is conceptually distinct from network connectivity; conflating the two is a common misconception in AI-900 exam preparation.

• ✗

  A training approach where users can interact with and correct the model in real time

  Why it's wrong here

  Human-in-the-loop correction, where users review and adjust model predictions in real time, is an interactive governance or active-learning mechanism, not an automated training technique. Online learning's weight updates are driven algorithmically by loss gradients computed from incoming labeled data, without requiring human intervention. Even if a user corrects a prediction instantly, the model may still update weights only in a separate batch cycle, so interactive correction does not inherently equate to continuous incremental weight updating.

Go deeper

Related to this question