ata
(Tensor<T>) where the scaled exponential linear unit function,
`y = gamma * (alpha * e^x - alpha) for x <= 0`, `y = gamma * x for x > 0`,
is applied to the tensor elementwise.