Generalization

Generalization is a model’s ability to perform usefully on relevant examples or conditions not seen during training.

Plan for this page

Understand one concept in plain language and connect it to nearby ideas.

You will leave with: A model generalizes when it has learned patterns that transfer beyond memorized training cases. The claim must name the target population, task, time, environment, and acceptable performance rather than assume universal transfer.
Time: 3–8 min
Before you begin: No prerequisite. Keep one real example in mind.
Do this now: Read the explanation, then compare the example with the common misunderstanding.

Prepared next stepExamine learning beyond examples

GeneralizationGeneralization is a model’s ability to perform usefully on relevant examples or conditions not seen during training.

Evaluation
Overfitting
Distribution shift

Plain explanation

A model generalizes when it has learned patterns that transfer beyond memorized training cases. The claim must name the target population, task, time, environment, and acceptable performance rather than assume universal transfer.

Example

A handwriting model trained on one set of writers generalizes if it also recognizes new writers whose samples fit the intended population.

Why it matters

Real users and situations differ from training data. Separate test data and deployment monitoring help reveal whether performance survives that change.

Common misunderstanding

Misconception: good test accuracy proves the model will work everywhere. Correction: it supports generalization only to conditions adequately represented by the test and method.

Related concepts

Follow relationships, not only the alphabet

EvaluationEvaluation is the structured measurement and investigation of how a model or system performs against defined questions, criteria, and contexts.OverfittingOverfitting occurs when a model fits training examples or their noise too closely and performs less well on relevant new cases.Distribution shiftDistribution shift is a meaningful difference between the data or conditions used to develop a model and those encountered later.

Examine learning beyond examples

Sources and scope

What supports this object

Machine Learning GlossaryGoogle for Developers

Defines generalization and related evaluation concepts; observed performance remains bounded by the tested data and conditions.

Open source

Loading page content...

…