Definition
The ability of a learned model or system to perform well on relevant examples, tasks, or environments not used to fit it.
Distinguish it from nearby terms
Generalization concerns transfer beyond fitted examples; benchmark performance measures only one sampled regime.
Check your understanding
A system may generalize within a benchmark family but fail under changed tools, budgets, or production conditions.