The figure below illustrates the various components of a perceptron -- the basic node in a neural network (also called "unit", or "neuron"). It has one or more connections feeding in, an equal number of weights, a bias scalar, an activation function, and an output connection. The perceptron multiplies each input by the corresponding weight, adds everything, adds bias, applies the activation function, and sends out the result. The activation is typically a non-linear function.