Chapter 2
Convolution by hand.
A convolution is one small calculation, repeated at every position in the image. This chapter does it by hand.
Chapter 1 ended with two properties a layer for images should have: each output reads only a small patch of neighboring pixels, and the same weights are used at every position. A convolution is the layer built from exactly those two ideas.
This chapter takes it apart. It starts with a single output: nine pixels, nine weights, one sum. Then it slides that calculation across the whole image to build an output map, looks at what different weights make the same slide find, and comes back to chapter 1’s moved duck to see what a convolution does with it.
Every calculation in this chapter can be done with pencil and paper, and the worked examples use small whole numbers so you can. The figures use the duck from chapter 1, drawn at 32 pixels on a side so its edges are easier to follow.
Modules
Each module ends in a short knowledge check. Passing it unlocks the next module. Your progress is saved in this browser.