Bias term
A constant a neuron adds no matter what, so it isn’t forced to pass through zero.
Think of it like
The base fare on a taxi meter — you pay it before the distance charge even starts.
Example
Without a bias, a neuron given all-zero input must output zero; the bias lets it say something else.
How it actually works
In y = Wx + b, the bias is b: a learnable offset that shifts the activation up or down independent of the input. Small thing, but it gives each neuron freedom to set its own baseline. Note: “bias” here is unrelated to fairness or social bias — same word, different sense.
For product teams
Pure mechanism; matters to you mainly as a reminder that model “bias” has two totally different meanings.
For engineers
The additive constant b in an affine layer, one learnable scalar per output unit.
Related
- Weights — The partner of weights.
- Weights and biases — Together they are weights and biases.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome