Deep learning with vector output

I need to learn a mapping from 28x28 images into a vector of 45 floating-point numbers. This is not really classification as the numbers range between -1 and 1.

When designing a deep neural network, what output layer could I use?

Best,

Samuli Siltanen

Answer 1

[Asvin Kumar]

You can use the tanhLayer to obtain output values in the range of –1 to 1.

Here’s the documentation for more information: https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.tanhlayer.html

[Samuli Siltanen]

Thank you for your answer! However, it seems that I cannot use tanhLayer as an output layer:

Error using trainNetwork (line 165)

Invalid network.

Caused by:

Network: Missing output layer. The network must have one output layer.

Layer 18: Unused output. Each layer output must be connected to the input of another layer.

[Asvin Kumar]

For the output layer, you can use a regressionLayer after the tanhLayer. This will produce predictions in the required range and compute the half-mean-squared-error loss.

Here's a link to the documentation: https://www.mathworks.com/help/deeplearning/ref/regressionlayer.html

[Samuli Siltanen]

Thank you so much! I will try this. Samu