Lesson 1 of 8

From linear models to neural networks

10 pts

Explanation

A single neuron is barely more than the linear model you already built: it takes inputs, multiplies each by a weight, adds a bias, then passes the result through an activation function (a nonlinear squashing function like sigmoid or ReLU). A neural network is many of these neurons arranged in layers, each layer feeding the next — that stacking of simple pieces is what lets networks represent very complex patterns.

Visual explanation

Read this from left to right: what goes in, what happens, and what comes out.

Input
Several layers
Complex prediction

Extra layers transform the input in stages before making the final prediction.

In plain language

A simple linear model makes one straight-style decision. Neural networks add layers so they can learn more complicated patterns.

Remember

More layers can learn richer patterns, but need more care and data.

What to do

Why is a single linear model not enough to represent complex patterns like recognizing handwritten digits, no matter how large you make the weights?

Validation questionWhat is missing from a purely linear model that layers of neurons with activation functions add?