torch (version 0.8.1)

nn_rrelu: RReLU module

Description

Applies the randomized leaky rectified liner unit function, element-wise, as described in the paper:

Usage

nn_rrelu(lower = 1/8, upper = 1/3, inplace = FALSE)

Arguments

lower

lower bound of the uniform distribution. Default: \(\frac{1}{8}\)

upper

upper bound of the uniform distribution. Default: \(\frac{1}{3}\)

inplace

can optionally do the operation in-place. Default: FALSE

Shape

• Input: \((N, *)\) where * means, any number of additional dimensions

• Output: \((N, *)\), same shape as the input

Details

Empirical Evaluation of Rectified Activations in Convolutional Network.

The function is defined as:

$$ \mbox{RReLU}(x) = \left\{ \begin{array}{ll} x & \mbox{if } x \geq 0 \\ ax & \mbox{ otherwise } \end{array} \right. $$

where \(a\) is randomly sampled from uniform distribution \(\mathcal{U}(\mbox{lower}, \mbox{upper})\). See: https://arxiv.org/pdf/1505.00853.pdf

Examples

if (torch_is_installed()) {
m <- nn_rrelu(0.1, 0.3)
input <- torch_randn(2)
m(input)
}