def __new__(cls, *args, **kwargs):
cls.name = "Restricted ImageNet"
cls.experiment_fn = 'restrictedImgnet'
grid_params = []
arch = "ResNet50"
grid_params.append({
'dataset': ['resImgnet112v3'],
'model': [
f'ce-tor-{arch}-adambs128',
f'advbeta2ce-tor-{arch}-adambs128',
f'advbetace-tor-{arch}-adambs128',
f'advbeta.5ce-tor-{arch}-adambs128',
f'stradesce-tor-{arch}-adambs128',
f'strades3ce-tor-{arch}-adambs128',
f'strades6ce-tor-{arch}-adambs128',
f'advce-tor-{arch}-adambs128',
f'sllr36ce-tor-{arch}-adambs128',
f'tulipce-tor-{arch}-adambs128',
],
'eps': [0.005],
'norm': ['inf'],
'attack': ['pgd'],
'random_seed': random_seed,
})
arch = "ResNet50_drop50"
grid_params.append({
'dataset': ['resImgnet112v3'],
'model': [
f'ce-tor-{arch}-adambs128',
f'advbeta2ce-tor-{arch}-adambs128',
f'strades6ce-tor-{arch}-adambs128',
f'strades3ce-tor-{arch}-adambs128',
f'advce-tor-{arch}-adambs128',
],
'eps': [0.005],
'norm': ['inf'],
'attack': ['pgd'],
'random_seed': random_seed,
})
arch = "ResNet50_drop20"
grid_params.append({
'dataset': ['resImgnet112v3'],
'model': [
f'ce-tor-{arch}-adambs128',
f'advbeta2ce-tor-{arch}-adambs128',
f'strades6ce-tor-{arch}-adambs128',
f'strades3ce-tor-{arch}-adambs128',
f'advce-tor-{arch}-adambs128',
],
'eps': [0.005],
'norm': ['inf'],
'attack': ['pgd'],
'random_seed': random_seed,
})
cls.grid_params = grid_params
return ExpExperiments.__new__(cls, *args, **kwargs)
def __new__(cls, *args, **kwargs):
cls.name = "mnist"
cls.experiment_fn = 'experiment01'
grid_params = []
arch = "CNN001"
grid_params.append({
'dataset': ['mnist'],
'model': [
f'advbeta2ce-tor-{arch}',
f'advbetace-tor-{arch}',
f'advbeta.5ce-tor-{arch}',
f'strades6ce-tor-{arch}',
f'strades3ce-tor-{arch}',
f'stradesce-tor-{arch}',
f'tulipce-tor-{arch}',
f'ce-tor-{arch}',
f'advce-tor-{arch}',
f'sllrce-tor-{arch}',
],
'eps': [0.1],
'norm': ['inf'],
'attack': ['pgd'],
'random_seed': random_seed,
})
arch = "CNN002"
grid_params.append({
'dataset': ['mnist'],
'model': [
f'advbeta2ce-tor-{arch}',
f'advbetace-tor-{arch}',
f'advbeta.5ce-tor-{arch}',
f'strades6ce-tor-{arch}',
f'strades3ce-tor-{arch}',
f'stradesce-tor-{arch}',
f'tulipce-tor-{arch}',
f'ce-tor-{arch}',
f'advce-tor-{arch}',
f'sllrce-tor-{arch}',
],
'eps': [0.1],
'norm': ['inf'],
'attack': ['pgd'],
'random_seed': random_seed,
})
cls.grid_params = grid_params
return ExpExperiments.__new__(cls, *args, **kwargs)
def __new__(cls, *args, **kwargs):
cls.name = "svhn"
cls.experiment_fn = 'experiment01'
grid_params = []
grid_params.append({
'dataset': ['svhn'],
'model': [
'ce-tor-WRN_40_10',
'advbeta2ce-tor-WRN_40_10',
'advbetace-tor-WRN_40_10',
'advbeta.5ce-tor-WRN_40_10',
'stradesce-tor-WRN_40_10',
'strades3ce-tor-WRN_40_10',
'strades6ce-tor-WRN_40_10',
'tulipce-tor-WRN_40_10',
#'advce-tor-WRN_40_10-lrem2',
'advce-tor-WRN_40_10',
'sllrce-tor-WRN_40_10',
#'aug01-ce-tor-WRN_40_10',
#'aug01-advbeta2ce-tor-WRN_40_10',
#'aug01-advbeta2ce-tor-WRN_40_10-lrem2',
#'aug01-advce-tor-WRN_40_10',
#'aug01-advce-tor-WRN_40_10-lrem2',
#'aug01-strades6ce-tor-WRN_40_10',
'ce-tor-WRN_40_10_drop50',
'advbeta2ce-tor-WRN_40_10_drop50',
'advce-tor-WRN_40_10_drop50',
'strades6ce-tor-WRN_40_10_drop50',
'strades3ce-tor-WRN_40_10_drop50',
#'aug01-ce-tor-WRN_40_10_drop50',
#'aug01-advbeta2ce-tor-WRN_40_10_drop50',
#'aug01-advce-tor-WRN_40_10_drop50',
#'aug01-strades6ce-tor-WRN_40_10_drop50',
#'aug01-strades3ce-tor-WRN_40_10_drop50',
],
'eps': [0.031],
'norm': ['inf'],
'attack': ['pgd'],
'random_seed': random_seed,
})
cls.grid_params = grid_params
return ExpExperiments.__new__(cls, *args, **kwargs)
def __new__(cls, *args, **kwargs):
cls.name = "cifar"
cls.experiment_fn = 'experiment01'
grid_params = []
grid_params.append({
'dataset': ['cifar10'],
'model': [
'ce-tor-WRN_40_10',
'tulipce-tor-WRN_40_10',
'stradesce-tor-WRN_40_10',
'strades3ce-tor-WRN_40_10',
'strades6ce-tor-WRN_40_10',
#'cure14ce-tor-WRN_40_10',
'advce-tor-WRN_40_10',
#'llrce-tor-WRN_40_10',
'sllrce-tor-WRN_40_10-lrem2',
],
'eps': [0.031],
'norm': ['inf'],
'attack': ['pgd'],
'random_seed':
random_seed,
})
grid_params.append({
'dataset': ['cifar10'],
'model': [
'aug01-ce-tor-WRN_40_10',
'aug01-tulipce-tor-WRN_40_10',
'aug01-stradesce-tor-WRN_40_10',
'aug01-strades3ce-tor-WRN_40_10',
'aug01-strades6ce-tor-WRN_40_10',
'aug01-advce-tor-WRN_40_10-lrem2',
#'aug01-scure14ce-tor-WRN_40_10-lrem4',
'aug01-llrce-tor-WRN_40_10',
#'aug01-sllrce-tor-WRN_40_10',
'aug01-sllrce-tor-WRN_40_10-lrem2',
],
'eps': [0.031],
'norm': ['inf'],
'attack': ['pgd'],
'random_seed':
random_seed,
})
cls.grid_params = grid_params
return ExpExperiments.__new__(cls, *args, **kwargs)
def __new__(cls, *args, **kwargs):
cls.name = "tiny ImageNet"
cls.experiment_fn = 'experiment01'
grid_params = []
arch = "ResNet152"
grid_params.append({
'dataset': ['tinyimgnet'],
'model': [
f'aug02-ce-tor-{arch}-bs256',
f'aug02-strades6ce-tor-{arch}-bs256',
f'aug02-advce-tor-{arch}-bs256',
f'aug02-llrce-tor-{arch}',
],
'eps': [0.031],
'norm': ['inf'],
'attack': ['pgd'],
'random_seed':
random_seed,
})
cls.grid_params = grid_params
return ExpExperiments.__new__(cls, *args, **kwargs)