Bias

Definition

Bias is an adjustable value added to a weighted sum.

Basic form:

score = w1*x1 + w2*x2 + b

Here:

b = bias

Bias is a learned offset.

---

Why Bias Is Needed

Weights control how inputs affect the score.

Bias shifts the score even when the inputs stay the same.

Example:

without bias:
score = 2*x

with bias:
score = 2*x + 5

The bias moves the output up or down.

Bias lets the neuron adjust its baseline.

---

Bias When Input Is Zero

Without bias:

x = 0
score = w*x
score = 0

With bias:

x = 0
score = w*x + b
score = b

The neuron can produce a nonzero score even when the input contribution is zero.

Bias gives the neuron a starting offset.

---

Bias as a Parameter

Bias is learned during training.

It is not an input.

It is part of the neuron's adjustable rule.

inputs change per example
bias changes through training

Optimization can increase or decrease the bias to reduce loss.

Bias is a trainable parameter, like a weight.

---

Weights vs Bias

Weights scale input values.

Bias shifts the result.

weight:
changes how much an input matters

bias:
changes the baseline score

Both help shape the neuron's behavior.

Weights control influence; bias controls offset.

---

Final Summary

• Bias is an added offset in a neuron.
• It shifts the weighted sum.
• Bias lets the neuron produce nonzero scores even when input contribution is zero.
• Bias is a trainable parameter.
• Weights scale inputs; bias shifts the result.
• Bias makes the neuron's transformation more flexible.