Long search time for samplewise perturbation with resnet18_madrys.yaml

[jasonchen505]

Hello,

I am running the code to find samplewise perturbations using the resnet18_madrys.yaml config on the CIFAR-10 dataset.

The problem is that the search process for both min-max and min-min perturbations runs for a very long time and does not seem to find a perturbation that satisfies the stop_error conditions. The settings in the config are: STOP_ERROR_minmin = 0.01,STOP_ERROR_minmax = 0.9.

Is this long search time expected for this specific configuration? Or is it recommended to modify these stop_error values to help the search converge successfully? Thank you for your help.