卷积神经网络的工作原理

How do Convolutional Neural

Networks work?

Data Science and Robots Blog

pdf [2MB] ppt [6MB]

in French by Charles Crouspeyre

Nine times out of ten, when you hear about deep learning breaking a new technological barrier, Convolutional Neural

Networks are involved. Also called CNNs or ConvNets, these are the workhorse of the deep neural network field. They have

learned to sort images into categories even better than humans in some cases. If there’s one method out there that justifies

the hype, it is CNNs.

What’s especially cool about them is that they are easy to understand, at least when you break them down into their basic

parts. I’ll walk you through it. There's a video that talks through these images in greater detail. If at any point you get a bit

lost, just click on an image and you'll jump to that part of the video.

X's and O's

To help guide our walk through a Convolutional Neural Network, we’ll stick with a very simplified example: determining

whether an image is of an X or an O. This example is just rich enough to illustrate the principles behind CNNs, but still

the images don’t match, at least to the computer. Ideally, we would like to be able to see X’s and O’s even if they’re shifted,

shrunken, rotated or deformed. This is where CNNs come in.

Features

CNNs compare images piece by piece. The pieces that it looks for are called features. By finding rough feature matches in

roughly the same positions in two images, CNNs get a lot better at seeing similarity than whole-image matching schemes.

Each feature is like a mini-image—a small two-dimensional array of values. Features match common aspects of the images.

In the case of X images, features consisting of diagonal lines and a crossing capture all the important characteristics of

most X’s. These features will probably match up to the arms and center of any image of an X.

Convolution

When presented with a new image, the CNN doesn’t know exactly where these features will match so it tries them