A sigmoid activation function is used in many neural networks. The properties of a sigmoid function mean that it shares some of the non-lonearity of a threshold function, but is differentiable enabling learning through the backpropagation algorithm.
Used in Chap. 6: page 87; Chap. 7: page 107; Chap. 9: page 137; Chap. 21: page 357
Sigmoid function smoothly limits extreme values