Convolutional neural network

General

A convolutional neural network consists convolutional, pooling and fully connected layers. They are used for example to classify images. Their advantage over regular neural networks is that they scale better with an increase in size of the input data. This is done by making certain assumptions over the input data.

Convolutional layer

In a convolutional layer the activity or output of neuron is calculated by applying a convolutional kernel to the corresponding input. The result of the kernel is the output the neuron.

Pooling layer

Pooling layers are used to reduce the size of the data, by scaling it down. This leads to a reduction in computational load.

Fully connected layer

In a fully connected layer each neuron of the previous layer is connected with all neurons on the next layer. They can be used to for example calculate the final class score, when classifying data.

Sources

• Convolutional neural network
• A Beginner's Guide To Understand Convolutional Nerual Networks
• Convolutional Networks