Steganografiya quyidagi sohalarda qo'llaniladi, lekin ular bilan cheklanmaydi

X_train (array): Modified input data
y_train (array): Modified label data

"""

if self.args.trigger_technique == 'noise':
ownership, X_train, y_train = self.generate_trigger_noise(
X_train,
y_train,
mode)
elif self.args.trigger_technique == 'dawn':
ownership, X_train, y_train = self.generate_trigger_dawn(
X_train,
y_train,
mode)
else:
raise NotImplementedError

return ownership, X_train, y_train

def generate_trigger_dawn(self, X_train, y_train, mode='CLASSIFICATION'):
"""Generation trigger set based on the paper.

DAWN: Dynamic Adversarial Watermarking of Neural Networks

by Szyller et al.

Args:
X_train (array): Input data
y_train (array): Label data
mode (str, optional): Classification or
regression mode

Returns:
ownership (dict): Watermark triggers information
X_train (array): Modified input data
y_train (array): Modified label data

"""
WM_X, WM_y = [], []
ownership = {}
for item, label in zip(X_train.to_numpy(), y_train):
if not self.__is_prediction_dawn(item):
WM_X.append(item)
WM_y.append(label)

ownership['inputs'] = WM_X
ownership['labels'] = WM_y
ownership['bounds'] = (min(y_train), max(y_train))

return ownership, X_train, y_train

def generate_trigger_noise(self, X_train, y_train, mode='CLASSIFICATION'):
"""Generation random trigger set.

Args:
X_train (array): Input data
y_train (array): Label data