TDNN Implementation in SNNS

The original time delay algorithm was slightly modified for implementation in SNNS, since it requires either variable network sizes or fixed length input patterns. Time delay networks in SNNS are allowed no delay in the output layer. This has the following consequences:

• The input layer has fixed size.
• Not the whole pattern is present at the input layer at once. Therefore one pass through the network is not enough to compute all necessary weight changes. This makes learning more computationally intensive.

The coupled links are implemented as one physical (i.e. normal) link and a set of logical links associated with it. Only the physical links are displayed in the graphical user interface. The bias of all delay units has no effect. Instead, the bias of the corresponding feature unit is used during propagation and backpropagation.

• Activation Function
• Update Function
• Learning Function