Deep learning generalizes because the parameter-function map is biased towards simple functions.

Guillermo Valle Prez, Chico Q. Camargo, Ard A. Louis

Browse the full ICLR paper archive.