Browse Teach Curate

Sigmoid Function

sigma(x) = 1/(1+e^-x) maps any real number into (0,1), which is why every LSTM gate uses it: its output acts as a soft, differentiable fraction of 'how much to let through'. Its gradient maxes out at 0.25 and saturates at both tails.

jeremy Video

Why Do We Use the Sigmoid Function for Binary Classification?

The sigmoid function is theoretically justified for binary classification by its unique saturation properties within cross-entropy loss landscapes, which ensure stable gradient propagation during opt…