Contents

BNNS.ActivationFunction.geluApproximation2(alpha:beta:)

An activation function that provides a fast evaluation of the Gaussian error linear units (GELU) approximation on its input.

Declaration

case geluApproximation2(alpha: Float, beta: Float)

Discussion

This constant defines an activation function that returns values using the following operation:

x * (ReLU 6(x + 3.0) * 1.0 / 6.0)

The following illustrates the output that the activation function generates from inputs in the range -10...10, an alpha of 0.1, and a beta of 1.0. The thinner, dashed line shows, for comparison, the result of BNNSActivationFunctionGELUApproximation using the same alpha and beta values:

[Image]

See Also

Activation Functions