Modifying loss function in neural network to be dependent on previous losses

Hi. I am training a special neural network and after each iteration I want to modify loss function so that it changes based on loss and fit from iteration right before. How could I do this? I do not need example of code that produces good results, just code thatd does that. A second question, is how do I close automautically graphs that open up while training in matlab, I have tried many solutions I believe would work so pleas eonly answer if you have succedded in this special case yourself.

[Torsten]

For your first question maybe "Dynamic Neural Networks" is the key term if "iteration" stands for "time":

https://uk.mathworks.com/help/deeplearning/ug/introduction-to-dynamic-neural-networks.html

If not, you should explain the reason behind the need to modify the loss function besed on iteration.

[Gad Licht]

Hi, I do nto think that works because its not an RNN, LSTM, or similar. More of a modified CNN. Time changes in loss function are also not dependent on data, but previous losses and fit.

Answer 1

[Matt J]

It would require a custom training loop,

https://www.mathworks.com/help/deeplearning/ug/train-network-using-custom-training-loop.html

[Gad Licht]

THank you.