the code needs to be updated slightly so that we can do a hybrid defense where we do rd, project back to the initial space, and then retrain again. It appears to be working in some code I have now, but I will need to retrain all of the models again. Doing it for the pca and n_epoch=50 first to check that it's working (i.e. that it's outperforming the recons defense for n_epoch=50), before proceeding with n_epoch=150. for pca, kpca; nd=100,200.
the code needs to be updated slightly so that we can do a hybrid defense where we do rd, project back to the initial space, and then retrain again. It appears to be working in some code I have now, but I will need to retrain all of the models again. Doing it for the pca and n_epoch=50 first to check that it's working (i.e. that it's outperforming the recons defense for n_epoch=50), before proceeding with n_epoch=150. for pca, kpca; nd=100,200.