Crate caffe2op_selu

source ·

Structs

  • | SeluGradient takes both Y and dY and | uses this to update dX according to the | chain rule and derivatives of the selu | function. |
  • | The Selu op takes one input tensor | $X$, an argument $alpha$, an argument | $scale$, and produces one output tensor | $Y$ of the same shape as $X.$ The op performs | the element wise Selu operation, | defined as | | $$y=selu(x) =\begin{cases}scale | (\alpha e^{x} - \alpha) & x < 0\scale | * x & otherwise\end{cases}$$ | | The default value of alpha is 1.6732632423543772848170429916717 | and the default value of scale is 1.0507009873554804934193349852946. | See Self-Normalizing Neural | | Networks | for more information. | | Github Links: | | - https://github.com/pytorch/pytorch/blob/master/caffe2/operators/selu_op.h | | - https://github.com/pytorch/pytorch/blob/master/caffe2/operators/selu_op.cc |