A nonmonotone learning rate strategy for SGD training of deep neural networks.

Nitish Shirish Keskar, George Saon

Browse the full ICASSP paper archive.