Convolutional Neural Networks: Layers

We learnt about the basic Convolutional Neural Net(CNN) architecture here. Now let’s see how we can add layers to this network so that it can be used for Computer vision applications.

This picture shows a CNN network with multiple layers. Each layer performs different operations. Let’s see how we build each layer.

There are 3 types of layers in a CNN. We will learn more about each of these layers below.

Let’s take a look at input/output of a Convolutional layer. In the images below: X or A = input parameters. W[subscript] is the parameter matrix for that layer and b is the bias for that layer.

The way we compute output of a layer is we apply non-linearity and add a bias to the input of convolution. The input of convolution could be the input image or it could be output of the previous layer.

Layers in Convolutional Network

Max pooling is usually used after a convolutional layer. Max pooling and convolution layer are considered to be part of the same layer L.

Next: Let’s take a look here at some classic CNNs that have been developed over the years.