How to use different transfer functions within the same layer of a neural network?

For example, hidden layer with 3 different neurons: 1.tansig, 2.logsig, 3.elliotsig. I understand that this influences normalization range and that training might not resault in good performance but i need it to test some other aspects. Thank you.

Answer 1

[Greg Heath]

If it were possible with the NNTBX, you would have to design a custom net.

The only way I can see doing it is to have 3 hidden layers that are only connected to the input and output but not to each other.

See the custom network section in the documentation.

Hope this helps.

Thank you for formally accepting my answer

Greg

[mladen]

Exactly this were my thoughts too. If and when I finish my work on this I'll post it here.