How a network computes, why gradients vanish, and what makes depth trainable
The sigmoid σ(x) = 1/(1 + e⁻ˣ) is the squashing function logistic regression already used in stats.26. It takes any real number and returns something strictly between 0 and 1, steeply near the middle and barely at all far out. For two decades it was also the default activation inside hidden layers. This lesson is about why it stopped being one.
🔒 This is a Pro lesson — the interactive figure, worked examples, quiz and practice open with Pro access.
▶ Sigmoid and Tanh: the Saturation Trap