An intercept, called a bias weight in a neural network, is an always-on input signal. Setting x0 to 1.0 lets w0 shift the neuron’s activation independently of the other inputs, just as b shifts linear regression output. section 5.1.5
Adding the bias to a perceptron
For token counts green=1, egg=2, ham=3, and spam=4, prepend 1 to the input vector and place w0 first in the weight vector. The weighted sum then includes w0 * 1.0 alongside the feature-weight products; a threshold converts the result into a prediction.
