Introduction

The first useful model family is linear.

A linear model computes a score from input features by multiplying, adding, and shifting with a bias. It is simple enough to compute by hand, but important enough to appear inside larger neural networks again and again.

This chapter studies the linear model as a model family:

score = x . w + b

or for many outputs:

scores = xW + b

By the end, you should be able to:

• compute a score by hand;
• explain weights and bias;
• read binary and multiclass scores;
• connect a score of zero to a decision boundary;
• explain why linear models are useful but limited.

The next chapters will add nonlinear activations and layers. Before that, the linear case should feel solid.