You Only Train Once: Loss-Conditional Training of Deep Networks.

Alexey Dosovitskiy, Josip Djolonga

Browse the full ICLR paper archive.