Module leaf::layers::activation::sigmoid
[−]
[src]
Applies the nonlinear Log-Sigmoid function.
Non-linearity activation function: y = (1 + e-x)-1
A classic choice in neural networks. But you might consider using ReLu as an alternative.
ReLu, compared to Sigmoid
- reduces the likelyhood of vanishing gradients
- increases the likelyhood of a more beneficial sparse representation
- can be computed faster
- is therefore the most popular activation function in DNNs as of this writing (2015).
Structs
Sigmoid |
Sigmoid Activation Layer |