Training a model once to make predictions across many similar problems, rather than solving each problem independently.