BigNet for Elman Networks

By clicking on the button in the BigNet menu, the BigNet window for Elman networks (see fig.) is opened.

  
Figure: The BigNet Window for Elman Networks

The number of units of each layer has to be specified in the column No. of Units. Each hidden layer is assigned a context layer having the same size. The values in the column No. of Col. have the same meaning as the corresponding values in the BigNet window for Jordan networks.

The number of hidden layers can be changed with the buttons and . adds a new hidden layer just before the output layer. The hidden layer with the highest layer number can be deleted by pressing the button. The current implementation requires at least one and at most eight hidden layers. If the network is supposed to also contain a context layer for the output layer, the button has to be toggled, else the button. Press the button to create the net. The generated network has the following properties:

• The layer i is fully connected to the layer i+1.
• Each context layer is fully connected to its hidden layer. A hidden layer is connected to its context layer with recurrent 1-to-1-connections.
• Each context unit is connected to itself.
• If there is a context layer assigned to the output layer, the same connection rules as for hidden layers are used.
• Default activation function for input and context units is the identity function, for hidden and output units the logistic function.
• Default output function for all units is the identity function

Click on the button to close the BigNet window for Elman networks.