Dropout

torch.nn.``Dropout(p=0.5, inplace=False)

During training, randomly zeroes some of the elements of the input tensor with probability p using samples from a Bernoulli distribution`.

• proven to be an effective technique for regularization and preventing the co-adaptation of neurons as described in the paper Improving neural networks by preventing co-adaptation of feature detectors .

[docs]class Dropout(_DropoutNd):
    def forward(self, input: Tensor) -> Tensor:
        return F.dropout(input, self.p, self.training, self.inplace)