[source] # ELU *Exponential Linear Units* are a type of rectifier that soften the transition from non-activated to activated using the exponential function. As such, ELU produces smoother gradients than the piecewise linear [ReLU](relu.md) function. $$ {\displaystyle ELU = {\begin{cases}\alpha \left(e^{x}-1\right)&{\text{if }}x\leq 0\\x&{\text{if }}x>0\end{cases}}} $$ ## Parameters | # | Name | Default | Type | Description | |---|---|---|---|---| | 1 | alpha | 1.0 | float | The value at which leakage will begin to saturate. Ex. alpha = 1.0 means that the output will never be less than -1.0 when inactivated. | ## Example ```php use Rubix\ML\NeuralNet\ActivationFunctions\ELU; $activationFunction = new ELU(2.5); ``` ## References [^1]: D. A. Clevert et al. (2016). Fast and Accurate Deep Network Learning by Exponential Linear Units.