Bridge To Probability And Loss

Where We Are

This chapter introduced calculus as the language of change.

The progression was:

function -> limit -> derivative -> gradient -> curvature

The main question was:

How does output change when input or parameters change?

Calculus gives a precise way to reason about change.

---

What Calculus Added

Calculus gave us:

derivative -> local rate of change
partial derivative -> one-input sensitivity
gradient -> multi-input sensitivity
chain rule -> change through composed functions
curvature -> change in local direction

These tools explain how a function responds to movement.

Calculus turns movement through a function into measurable structure.

---

Why Change Is Not Enough

Knowing how a function changes is only part of learning.

We also need to know:

what the model is trying to predict
how wrong the output is
how uncertainty is represented
what better means

Those questions require probability and loss.

Change tells how to move; loss tells what movement should improve.

---

Why Probability Comes Next

Many learning systems do not only output a single hard answer.

They often represent uncertainty.

Example:

class A: 0.8
class B: 0.2

This is not just a decision.

It is a distribution over possible outcomes.

Probability gives language for uncertainty.

---

Why Loss Matters

A loss function measures error.

Basic idea:

prediction + target -> loss

Low loss means the prediction is closer to the target under the chosen rule.

High loss means the prediction is worse.

Loss turns wrongness into a quantity.

---

How Calculus Connects To Loss

Once loss is a function, calculus can ask:

How does loss change when a parameter changes?

That leads to gradients of loss.

Basic shape:

parameters -> prediction -> loss

The chain rule tracks how parameter changes affect the loss through the prediction.

Calculus becomes useful for learning when error is expressed as a function.

---

Bridge Question

The next chapter begins with a new question:

How do we represent uncertainty and error?

That leads to:

random variables
probability distributions
expectation
variance
conditional probability
likelihood
model output as distribution

Probability describes uncertainty; loss gives learning something to reduce.

---

Final Summary

• Calculus describes change in functions.
• Derivatives and gradients measure local sensitivity.
• Curvature describes local shape beyond slope.
• Learning also needs a way to describe uncertainty and error.
• Probability handles uncertainty.
• Loss turns prediction error into a quantity.
• The next chapter moves from change to uncertainty.