Convolutional Networks
A photo is a big grid of numbers. A convolutional network (CNN) is a neural network built to read that grid. It is the idea behind phone cameras that find faces, and tools that read number plates. In this lesson you slide a small filter over a picture and see what it finds.
New words
- Pixel
- One tiny dot of a picture, stored as a number for its brightness.
- Filter (kernel)
- A small grid of weights that is slid over an image.
- Convolution
- Multiply a filter with a patch of the image, then add the results.
- Feature map
- The grid of results you get after sliding one filter over an image.
- Stride
- How many pixels the filter moves each time.
- Padding
- Extra zeros added around the image so the filter can reach the edges.
- Pooling
- Shrinking a map by keeping one value from each small block.
What you will learn
- How does a computer see a picture?
- Why not use a normal network?
- What is a filter?
- Convolution in code and edge filters
- Stride, padding and pooling
- Try it: slide a filter over a picture
- Stacking layers: edges, shapes, objects
- Common mistakes
- Explain it to a friend
- Key takeaways
10 quiz questions at the end.
The full lesson, with interactive demos and the quiz, opens after you sign in with Google. It is free.