Bias Term

A learnable parameter in a neural network node that shifts the activation function, allowing the model to fit data that does not pass through the origin. It gives the model more flexibility to represent a wider variety of patterns.
See also: activation function, weight.