How to create a simple fully connected neural network with multiple outputs?

I need to create a fully connected neural network that can have multiple otputs.

I see RegressionNeuralNetwork is a very good solution for me, but its output size can only be 1.

https://www.mathworks.com/help/stats/regressionneuralnetwork.html

Please refer me to an example.

Answer 1

[Ashu]

Hey Mahmoud,

To train a network with multiple outputs, you must train the network using a custom training loop.

Example on Training and Inferencing Multiple Output Neural Network : https://www.mathworks.com/help/deeplearning/ug/train-network-with-multiple-outputs.html

To understand more about Multiple Input and Output Neural Networks : https://www.mathworks.com/help/deeplearning/ug/multiple-input-and-multiple-output-networks.html

Regards

[Mahmoud Elzouka]

Thanks @Ashu for your answer.

I would like to "create" the NN from known parameters (i.e., biases and weights). Would you please share an example?

[Ashu]

Hey Mahmood,

To set the weights and biases, you can use 'setwb'.

Here is a small example of creating a network with multiple outputs :

x = randn(18,141); % input data

t = randn(18,141); % ground truth label

net = feedforwardnet([ 36 36 ]);

net = train(net,x,t);

view(net)

Now to set the weights and biases -

net = setwb(net,rand(10,1));

To view the parameter values-

net.IW{1,1}

net.b{1}

To know more about 'setwb' you can refer this -

https://www.mathworks.com/help/deeplearning/ref/setwb.html