Suivre

Very interesting paper, which says that you can provably make all local optima global by just adding a neuron next to the output neurons of a model !

It's yet another one in a series of papers that give partial explanatory theories about why deep learning works so well.

If anyone has the time to check the proofs, please let me know.

arxiv.org/abs/1901.00279

Inscrivez-vous pour prendre part à la conversation
Mastodon - etalab

Instance ouverte pour test - lire les CGU Elle est ouverte à tout agent possédant un compte email en ".gouv.fr" ou la liste disponible sur https://forum.etalab.gouv.fr/t/mastodon-le-reseau-social-libre-et-decentralise-en-plein-decollage/3538