Weight decay
Also: L2 regularisation
Shrinking every weight slightly towards zero at each step, so the model only keeps magnitude it is actually using.
A definition is not an understanding. This is taught properly in Batch size, schedules and regularisation in practice, part of Backpropagation.
Learn this properlyTerms from the same topic