Introduction

A model is a function with parameters.

That sentence is short, but it carries most of deep learning.

The input changes from example to example. The parameters are learned values stored inside the model. The output is the model's prediction for that input. During one forward pass, the model only computes. During training, the parameters are adjusted so future predictions improve.

This chapter makes that separation precise.

By the end, you should be able to explain:

• what enters a model;
• what the model stores;
• what a forward pass computes;
• why prediction comes before learning;
• what changes during training;
• why model capacity means a family of possible functions.

We will still avoid frameworks. The point is to make the model visible before a library compresses it into a class or module.