Modify Loss Function in Predefined Network

Hi All--I'm relatively new to deep learning and have been trying to use a large data set to predict cardiovascular disease between healthy/sick subjects. I have been trying to train existing networks (AlexNet, etc.). The issue is that my control data vs sick data is about 80%/20% split, so I would like to penalize the loss function for calling a sick patient not sick. This is where I am stuck. Here is my last layer:

lgraph.Layers(25)

ans =

ClassificationOutputLayer with properties:

Name: 'new_classoutput'

Classes: 'auto'

OutputSize: 'auto'

Hyperparameters

LossFunction: 'crossentropyex'

How do I modify the loss function within this layer? Thanks a bunch!

Answer 1

[Harsha Priya Daggubati]

Hi,

You can try creating your custom loss function. Refer to the following link:

https://www.mathworks.com/matlabcentral/answers/462440-custom-loss-function-for-dnn-training