Hands-On Reinforcement Learning with Python

Fully connected layer

We can have multiple convolutional layers followed by pooling layers. However, these layers will only extract features from the input image and produce the activation maps. How do we classify whether there is a dog in the image with the activation maps alone? We have to introduce a new layer called a fully connected layer. It receives input as the activation maps (which are now basically the features of the image) apply the activation function and it produces the output. A fully connected layer is actually the normal neural network where we have an input layer, hidden layer, and output layer. Here, instead of an input layer, we use convolution and pooling layers, which together produce the activation maps as an input. ...