How do we decide the number of units in the hidden layer in a Time Delay Neural Network. I know that a regular feed-forward network uses Kolmogorov's principle and has 2d+1 units in the hidden layer. But that doesnt apply to TDNNs. Does anyone know of a similar method or any literature applicable for TDNNs??
Well, my 2 bits is that there is no clear (almost no) theoretical models for this. There are general heuristics which people use. And then when people use a new dataset and different cleaning techniques, they find different settings.
I'm dealing with 20 dimensional data extracted EEG signals. Any suggestions with the network design?
you should use trial and error technique. You should loop over different numbers of neurons in the hidden layer then choose the one that gives you the best result.
It's not true that the number of units in a hidden layer is determined independently of the association studied. The number of hidden units-and the number of hidden layers more generally-depends on properties of the association.
Hit and trail work the best for setting the number of neurons in the hidden layer of ANN.
These are no generalized rule to select the number of neurons in the hidden layer. It varies from problem to problem. The best way to select the optimum number of neurons is to do repeated testing by changing the number of neurons.
Maybe this theoretical work helps:
Article On the approximation by neural networks with bounded number ...
- Badges
- Science topic