BigNet for Self-Organizing Maps

As described in chapter , it is recommended to create Kohonen-Self-Organizing Maps only by using either the BigNet network creation tool or convert2snns outside the graphical user interface. The SOM architecture consists of the component layer (input layer) and a two-dimensional map, called competitive layer. Since component layer and competitive layer are fully connected, each unit of the competitive layer represents a vector with the same dimension as the component layer.

 
Figure: The BigNet window for the SOM architecture

To create a SOM, only 3 parameters have to be specified:

• Components: The dimension of each weight vector. It equals the number of input units.

• X-size: The width of the competitive layer. When learning is performed, the x-size value must be specified by the fifth learning parameter.

• Y-size: The length of the competitive layer. The number of hidden (competitive) units equals X-size Y-size.

If the parameters are correct (positive integers), pressing the button will create the specified network. If the creation of the network was successful a confirming message is issued. The parameters of the above example would create the network of figure . Eventually close the BigNet panel by pressing the button.

  
Figure: An Example SOM