how to define the size of feedback delays in Layer recurrent neural networks?

How can I find the right size of my feedback delays in a LRNN ?

Answer 1

[Greg Heath]

I DON'T KNOW.

So, I would start with using the same input and feedback lags I would use for NARXNET

 input: Significant lags of the training set input-target cross correlation function

 feedback: Significant lags of the training set target auto correlation function

Hope this helps.

Thank you for formally accepting my answer

Greg

[Mohamed Amine Chaaleb]

thanks for the answer Greg! I have a layer recurrent neural network with multiple outputs : let's say Output1,Output2 and Output3. When it comes to the correlation to determine the feedback delay do I have to determine the autocorrelation or the crosscorrelation. which signals do I have to corrolate ? Every Output with itself in means of nncorr(Outpu1,Output1) or do I even have to corrolate for example nncorr(Output1,Output2) because of the multiple Outputs? I mean here to determine the feedback delay. For the Input delay I'm going to crosscorrolate Input1 with Output1 and then Input1 with Output2 and then Input2 with Output1 etc ... I'm I right ?