Activation Functions - SkyWorld117/DianoiaML.jl GitHub Wiki

Functionality

The activation functions change the features of output data from each layer. The current framework supports None, ReLU, Sigmoid, Softmax and tanH. They are modules in Julia.

tanH

In order to avoid using the same name of Julia built-in function tanh, the name is slightly changed to tanH.

Softmax

To avoid the numerical instability, please use Softmax_CEL if you want to use Softmax with the loss function Cross_Entropy_Loss.