Research Slogan

Learn patterns from the training set and validate their generalization on a held-out set. When a pattern holds on unseen data, that is evidence of a generalizable principle rather than mere overfitting.