Data augmentation
Expanding the training set with transformed copies of examples that keep the same label, so the model learns invariances instead of memorizing specifics. Common forms include image flips, crops, color jitter, MixUp, and synonym swaps or back-translation for text.
Example
One 32×32 photo of a cat becomes many training images by mirroring it, cropping random 28×28 patches and rotating by up to ±15°; every copy keeps the label "cat".