## The Hidden Layer You do not have to read or understand this section unless you want to!

When it is being trained to recognize a font a Scan2CAD neural network is made up of three parts called “layers” – the Input Layer, the Hidden Layer and the Output Layer. See Advanced neural network information for a diagram.

The Hidden Layer is the part of the neural network that does the learning. It takes example characters from the Input Layer and learns to match them up with the characters you are training Scan2CAD to recognize, which are listed in the Output Layer.

## Nodes

The Hidden Layer contains “nodes” (these are different from the nodes in the Input Layer and the Output Layer). By default there is one node for each character you are trying to teach the neural network to recognize.

To take the letter B as an example: As you present example B shapes to the B node, the B node builds up an image of what a B is likely to look like.

If the example B shapes are all quite similar, each example B will reinforce the B node’s idea of what a B looks like so that the neural network will be more likely to recognize a B accurately in future.

If however there is wide variety in the example Bs being presented to the B node it will be less certain about what exactly constitutes a B and will therefore be less accurate in its recognition and less likely to be able to discriminate between a B and a similar character such as an 8.

For this reason it is often better to train one neural network to recognize one font and then train another neural network to recognize another font rather than try to train one neural network to recognize everything.

## The number of nodes in the Hidden Layer

It is possible to have more than one node in the Hidden Layer for each character you are trying to recognize. The suggested maximum number of nodes is:

No. of nodes in the Input Layer x No. of nodes in the Output Layer

## Advantages of increasing the number of nodes in the Hidden Layer

Increasing the number of nodes in the Hidden Layer can help the neural network to recognize variations within a character better. For example, one B node could learn to recognize tall thin Bs and another B node could learn to recognize short wide Bs.

## Disadvantages of increasing the number of nodes in the Hidden Layer The neural network may take significantly longer to train. You will need more example characters so that there will be enough examples of the different types of B to allow the neural network to learn the common features of each type. General character recognition may worsen. The neural network may become very good at recognizing the specific B shapes in the training set, but because it has learnt so many varieties it may not have learnt the common features that define a B. So when confronted with an example of a B it hasn’t seen before it may be unable to recognize it.

## Edit the number of nodes in the Hidden Layer

You can edit the number of nodes in the Hidden Layer when you initialize a neural network.

 1 Select Train Menu > Initialize Neural Net.
 2 Click on the Advanced button.

The Initialize New Training Net dialog appears.

 3 Edit the number of nodes using the slider control. 4 Click OK.

## View the number of nodes in the Hidden Layer

To view the number of nodes in the Hidden Layer select Train Menu > Inspect Neural Net.

Have questions on this topic?