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.

• ✗

  Making the model work for all programming languages and platforms

  Why it's wrong here

  This describes cross-platform or polyglot software engineering concerns, such as deploying a model across different programming languages, operating systems, or hardware architectures. Generalization is not about the model's compatibility with various technical environments but about its statistical behavior on data it has never encountered. A model that generalizes well in one deployment environment may still require significant engineering to run on multiple platforms; conversely, a model can be made cross-platform while memorizing training data and failing on new inputs. Thus, platform compatibility is an infrastructure detail, not a learning property.

• ✓

  The model's ability to perform well on new, unseen data by learning underlying patterns rather than memorizing training examples

  Why this is correct

  This is the correct definition: generalization is the model's ability to apply learned patterns, rather than memorized training examples, to new, unseen data. When a model memorizes the training set (overfitting), it may achieve near-perfect training accuracy but poor performance on real-world inputs, because it fails to capture the underlying regularities in the data. Generalization is typically measured by evaluation on a held-out validation or test set, and techniques such as regularization, dropout, and data augmentation are used to improve it. This concept is fundamental to supervised machine learning, as the ultimate goal is to make accurate predictions on future, previously unseen examples.

• ✗

  Making the model output descriptions in plain language for non-technical users

  Why it's wrong here

  This describes explainability or interpretability, where a model's predictions are communicated in human-readable language so users can understand the reasoning. Generalization, by contrast, is an empirical property of model performance: how well it maps novel inputs to correct outputs based on patterns learned during training. A model can generalize extremely well yet still produce opaque outputs, and conversely, a model can provide plain-language explanations while performing poorly on new data. Therefore, altering output communication style does not define generalization.

• ✗

  Training a model that works for all possible tasks without specialization

  Why it's wrong here

  This describes a universal model capable of handling any conceivable task, which corresponds to artificial general intelligence (AGI) or very large multi-task foundation models, not the narrower concept of generalization. Generalization specifically means that a model trained for a particular task (e.g., image classification) performs accurately on new, unseen examples from the same task distribution. A model that works for all possible tasks would require reasoning and adaptability far beyond standard machine learning, whereas generalization is about how well learned patterns transfer within the model's designated problem domain.

Go deeper

Related to this question