Is this overfitting?

1

This is my training and validation loss on 5-fold CV for a classification task. Is this overfitting ?

I’m using Focal Loss, AdamW(lr=1e-3, weight_decay=0.01), and ReduceLROnPlateau(factor=0.5, patience=3). What I can do to make validation loss more stable (another optimizer or lr_cheduler or anything) ?
Thank you !

loss_fold1 loss_fold2 loss_fold3 loss_fold4 loss_fold5

2

The loss curves indicate overfitting, but it also seems that not many samples might be used since the curves are showing some “discrete” steps.

In order to counter overfitting you could add more regularization to the model, e.g. through dropout, weight decay etc.

1 Like