Abstract: Batch normalization (BN) allows training very deep networks by normalizing activations by mini-batch sample statistics which renders BN unstable for small batch sizes. Current small-batch ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results