IDropout instances operate on an activations array, modifying or dropping values at training time only.
AlphaDropout is a dropout technique proposed by Klaumbauer et al.
Implements standard (inverted) dropout.
Regarding dropout probability.
Applies additive, mean-zero Gaussian noise to the input - i.e., x = x + N(0,stddev).
Note that this differs from
Note also that schedules for the standard deviation value can also be used.
Spatial dropout: can only be applied to 4D (convolutional) activations.