Layers in artificial neural networks

Before we go any further, its worth reminding ourselves about how ‘regular’ artificial neural networks work. You will have encountered these already if you have completed the preceding course in this series Machine Learning for Image Data.

Layers in artificial neural networks

In brief, artificial neural networks consist of a series of sequentially arranged layers, each consisting of a fixed set of nodes or neurons, each of which is connected to every neuron in the directly preceding and following layers.

The numerical data signal is only passed in one direction, from the input layer to the output layer, and the amount of signal passed through each connection between neurons is determined by the weight (a number) attached to each individual connection in the network.

By adjusting these weights the network can be trained to give a particular desired output (or answer) to a given piece of input data.

A layer in a conventional artificial neural network is the set of neurons positioned at the same distance through the network, plus also all the weights connecting it to the next layer in the network.

So, if for example there are 64 neurons in one layer, and 32 neurons in the following layer, there will be 64 x 32 = 2048 weights associated with that layer. Seems a lot, but wait until we see some convolutional layers!

This kind of layer is still found within convolutional neural networks and is known as a fully connected layer.

In the next article we will see how convolutional layers differ from these fully connected layers.

Image (c) The University of Nottingham