First-order ANIL provably learns representations despite overparametrisation.

Oguz Kaan Yksel, Etienne Boursier, Nicolas Flammarion

Browse the full ICLR paper archive.