How to create Multiple Output Neural Networks

Hi,

I am using the custom neural network maker network () to create my own architecture. However, I want multiple outputs to come from one layer (the last layer) but I am only able to make one output from the net. Is there anyway to do this?

I dont want to make a new network for each layer.

Kind regards,

E

Answer 1

[Greg Heath]

It is very simple:

For N I-dimensional "I"nputs paired with N O-dimensional "O"utput targets

 [ I N ] = size(inputmatrix)

 [ O N ] = size(targetmatrix)

Hope this helps.

Thank you for formally accepting my answer

Greg

[Emilio Lemonaris]

Hi Greg, thank you for your reply.

I dont quite understand how i can implement this using the matlab function 'network' when the output variable is a read-only value, it cannot be assigned dimension.

E

[Greg Heath]

The target matrix contains N columns of O-dimensional vectors with nonnegative unit sum components that can be interpreted as conditional (on the input) probabilities.

Each of the O-dimensions represent an output category.

Hope this helps.

Greg

[Emma Reeves]

Greg,

Is it possible to set up a structure where

[I N1] = size(inputmatrix);

[O N2] = size(outputmatrix);

N1 /= N2?

I know that other programs such as TensorFlow can do it. I don't understand why the second dimension needs to match.

[Greg Heath]

NO!

EACH of the N O-dimensional "O"utput target column vectors is the desired output for ONE of the N I-dimensional "I"nput column vectors.

Why?

Because that is the MATLAB NN convention regardless if it's pattern-recognition/classification or curve-fitting/regression.

EXAMPLE: For two-class classification, the targets are either [1 0 ]' or [0 1 ]'. However, if you have 60 inputs from class 1 and 40 from class 2, you still need 100 target columns.

Hope this helps.

Greg

Answer 2

[amirhossein boniadi]

Hi;

I want to find out the algoritm used to coding 8 bit digital input to 8 bit digital out put can i find it by using neural networks in matlab? and can any one give me a brief instruction...

thanks...