Matlab Neural Network toolbox: How to define one-by-one connection between neurons of two subsequent layers?

I have designed a network using Matlab Neural Network toolbox.

The input vector ( net.inputs{1}) has 100 elements and the first intermediate layer ( net.layers{1}) has also 100 neurons. ( net.layers{1}.size=100 )

The weight matrix between these two ( inputWeight{1,1} ) has the size of [100 100] which means the default connection shape is fully connected.

How can I change it in such to have a one-by-one connection between each input element and each neuron of the next layer? In other words how can I change the inputWeight{1,1}.size to [100 1]?

thanks,

P.S.: As you know, inputWeight{1,1}.size is read-only and depends on layers' definition. So, I cannot change it explicitly.

Answer 1

[Darshan Ramakant Bhat]

As of now MATLAB supports only fully connected layers. However there are workarounds possible. Please go through the following link dealing with the similar issue

http://www.mathworks.com/matlabcentral/newsreader/view_thread/33807

For more information on how to build customize Neural Network Architectures please refer following article

https://www.mathworks.com/help/nnet/ug/create-and-train-custom-neural-network-architectures.html

I hope this will help.

Regards

Darshan Bhat

[Robert Macrae]

Issue remains interesting, so please could you update the broken newsgroup link?