Neural Network Regression Problem with multiple Outputs

Hello everyone, I am now having a 1000 images, which have 10 continous attributes. My goal is to predict these 10 attributes values when I feed it a new image, after training.

So I think it is like a regression problem and I expect there will be 10 regression output layers in respond to these 10 attributes. How can i achieve this with Matlab? I think trainNetwork doesnt work. It is only applicable for non-multi-output lgraph.

Then I look at this:Train Network with Multiple Outputs https://www.mathworks.com/help/deeplearning/examples/train-network-with-multiple-outputs.html

Is that right direction? If not, can you point me?

I dont really know what parameters and state are.

I would be gladful if you leave a comment.

Answer 1

[Srivardhan Gadila]

The last two layers of your network architecture must be a fullyConnectedLayer with outputSize 10 followed by regressionLayer

layers = [ ...
    fullyConnectedLayer(10) 
    regressionLayer];