What is the significance of the tapped delay lines in dynamic neural networks?

What are the effects they have on network performance and how do I know what values of delays (input and layers) to choose for a particular application ?

Answer 1

[Greg Heath]

ID >= 0 causes the output to depend on current and past inputs

FD > 0 causes the output to depend on past outputs

Significant nonnegative input delays can be obtained from the significant peaks in the input/target cross-correlation function

Significant positive feedback delays can be obtained from the significant peaks in the target autocorrelation function

For examples, search the NEWSGROUP and ANSWERS for the most recent posts obtained from the search words

 greg nncorr

The correlation functions can also be obtained from the fft , and other tool boxes (Search the mathworks website).

Hope this helps

Thank you for formally accepting my answer

Greg

[Kingsley Akpeji]

Thanks Greg. But, the part of your answer talking about obtaining significant non-negative input delays and feedback delays is a little vague to me.

I don't know, but I suppose you are really great with the Neural Network toolbox. I really need help in applying it for a part of my undergraduate final year project. Can I get your email, so I can give you an explicit description of what I am doing, probably you can help me somehow.

[Greg Heath]

I only communicate via NEWSGROUP CENTRAL or ANSWERS