How to build a neural network which is not Fully-connected with NN toolbox?

Hi, I'm using NN toolbox to build my own network. The problem is that it seems that NN toolbox offers only fully-connected network. The image attached can be one example. Is there any way that I can build a neural network with disconnecting some weights?

[Itay Hanoch]

Hi,

I run into the same problem as you,

Trying to find a way to disconnect specific weight in a layer,

Did you you found a way to deal with this problem at the end?

Answer 1

[Greg Heath]

The best approach is to find, via an exhaustive search within bounds, the minimum number of hidden nodes that will yield your desired result.

I have posted ZILLIONS of examples in both the NEWSGROUP (comp.soft-sys.matlab) and ANSWERS.

For

 1. N I-dimensional "I"nputs yielding N O-dimensional "O"utputs
 2. Default 0.7/0.15/0.15 data division
 3. H hidden units in a default I-H-O node topology

 Ntrneq ~ 0.7*N*O         % No. of training equations
 Nw    = (I+1)*H+(H+1)*O  % No. of unknown weights

Find the minimum number of hidden units that will guarantee

   Ntrneq >= Nw

or

   H  <= (Ntrneq-O)/(I+O+1)

subject to the following target variance performance constraint on the error

 error = target-output

 mse(error) <= 0.01*var(target',1)

Hope this helps.

Thank you for formally accepting my answer

Greg

[Dong gun Lee]

I'm not sure that you correctly understood my question. As a default, it seems that MATLAB NN toolbox only offers fully connected network. I want to build a network which is not fully-connected. I already fixed the number of neurons in input and hidden layer. The only thing which should be solved is to disconnect some connections.