RReLU

class torch.nn.RReLU(lower=0.125, upper=0.3333333333333333, inplace=False)

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

Empirical Evaluation of Rectified Activations in Convolutional Network.

The function is defined as:

$$\text{RReLU}(x) = \begin{cases} x & \text{if } x \geq 0 \\ ax & \text{ otherwise } \end{cases}$$

where $a$ is randomly sampled from uniform distribution $\mathcal{U}(\text{lower}, \text{upper})$ during training while during evaluation $a$ is fixed with $a = \frac{\text{lower} + \text{upper}}{2}$.

See: https://arxiv.org/pdf/1505.00853.pdf

Parameters

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

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

• inplace (bool) – can optionally do the operation in-place. Default: False

Shape:

• Input: $(*)$, where $*$ means any number of dimensions.

• Output: $(*)$, same shape as the input.

Examples:

>>> m = nn.RReLU(0.1, 0.3)
>>> input = torch.randn(2)
>>> output = m(input)