Bias

A bias is an additive learned parameter that shifts a model's activation curve left, right, up, or down independently of the weighted input contributions, allowing the model to fit data that does not pass through the origin.

Real-World Mental Model

“The baseline water pressure in a faucet when nobody is turning the handle.”

INTERACTIVE ENGINE: BIAS SLIDER

Live Reactive

Bias Parameter (b): Intercept Shift0.0

-4.0 (Suppresses Activation)0.0 (Neutral)+4.0 (Triggers Activation)

Threshold Equation: z = (2.0 × 1.0) + (0.0) = 2.0STATE: FIRED (1)

Key Mental Model: Bias enables the neuron to output non-zero values even when all input features x are exactly zero, shifting the hyper-plane decision boundary across space.

CORE PRINCIPLES

What You Must Understand

Without bias, a linear neuron must pass through the origin (0,0) when all inputs are 0.

Acts as a learned activation threshold: determines baseline firing likelihood.

Distinct from sociological or demographic bias.

Reference Implementation

// Linear Neuron Equation
const z = (weight * x) + bias;

Production Failure Mode

Forgetting to include a bias term, forcing linear decision boundaries to intersect (0,0) and severely restricting representational capacity.

View Full 26-Topic Curriculum Table

Loading AI Engineering Foundations...