Accelerated training through iterative gradient propagation along the residual path.

Erwan Fagnou, Paul Caillon, Blaise Delattre, Alexandre Allauzen

Browse the full ICLR paper archive.