Viet-Anh on Software Logo

What is: Tanh Activation?

Data SourceCC BY-SA -

Tanh Activation is an activation function used for neural networks:

f(x)=exexex+exf\left(x\right) = \frac{e^{x} - e^{-x}}{e^{x} + e^{-x}}

Historically, the tanh function became preferred over the sigmoid function as it gave better performance for multi-layer neural networks. But it did not solve the vanishing gradient problem that sigmoids suffered, which was tackled more effectively with the introduction of ReLU activations.

Image Source: Junxi Feng