Deep Learning Toolbox: Inputs and Outputs

Hi everyone, I have started to transition from my own Matlab ML tool suite into the Deep Learning toolbox, and I have been unable to do the most simple of problems (a vector of elements for an input and a vector of elements for an output). Do you all know how to do this with mulitple hidden layers?

I have tried two different paths, and so far, I have failed royally.

For the 'trainNetwork', I am looking for an input layer for something that isn't a matrix in time sequence or not, but I am able to output a vector of outputs.

For 'train', I have been able to figure out how to define multiple inputs, but I am unable to output more than one value from the final layer.

This is rather dumbfounding, and I hope one of you have figured out how to fix this problem.

The image below may help you all grasp what I am trying to do, and I should note that this will be integrated into an AutoML process to define the network structure, so a GUI based answer won't work.

Matlab 19a with Deep Learning Toolbox, but I could transfer to 19b (I believe) if required.

Answer 1

[Blake Van Winkle]

If you are having this issue, I found that it was much easier to just use imageInputLayer but not have it defined as an image format.

For a vector: imageInputLayer([n,1,1])

For a 2d matrix: imageInputLayer([n,m,1])

For a 3d matrix: imageInputLayer([n,m,o])

Thanks for other suggestions...

Answer 2

[Sai Bhargav Avula]

Hi,

If the network also has multiple inputs and outputs, then you must define the network as a function and train the network using a custom training loop. for more information.

For a detailed understanding on how to it needs to be done refer the following link.

https://www.mathworks.com/help/deeplearning/ug/multiple-input-and-multiple-output-networks.html

Hope this helps

[Blake Van Winkle]

To Sai:

While I am assuming the documentation makes sense to some, it is clear as mud to me. You wouldn't happen to have an example handy?

Thanks