mindarmour.adv_robustness.defenses

This module includes classical defense algorithms in defencing adversarial examples and enhancing model security and trustworthy.

class mindarmour.adv_robustness.defenses.AdversarialDefense(network, loss_fn=None, optimizer=None)

Adversarial training using given adversarial examples.

Parameters

• network (Cell) – A MindSpore network to be defensed.

• loss_fn (Functions) – Loss function. Default: None.

• optimizer (Cell) – Optimizer used to train the network. Default: None.

Examples

>>> class Net(Cell):
>>>     def __init__(self):
>>>         super(Net, self).__init__()
>>>         self._reshape = P.Reshape()
>>>         self._full_con_1 = Dense(28*28, 120)
>>>         self._full_con_2 = Dense(120, 84)
>>>         self._full_con_3 = Dense(84, 10)
>>>         self._relu = ReLU()
>>>
>>>     def construct(self, x):
>>>         out = self._reshape(x, (-1, 28*28))
>>>         out = self._full_con_1(out)
>>>         out = self.relu(out)
>>>         out = self._full_con_2(out)
>>>         out = self.relu(out)
>>>         out = self._full_con_3(out)
>>>         return out
>>>
>>> net = Net()
>>> lr = 0.0001
>>> momentum = 0.9
>>> loss_fn = SoftmaxCrossEntropyWithLogits(sparse=True)
>>> optimizer = Momentum(net.trainable_params(), lr, momentum)
>>> adv_defense = AdversarialDefense(net, loss_fn, optimizer)
>>> inputs = np.random.rand(32, 1, 28, 28).astype(np.float32)
>>> labels = np.random.randint(0, 10).astype(np.int32)
>>> adv_defense.defense(inputs, labels)

defense(inputs, labels)

Enhance model via training with input samples.

Parameters

• inputs (numpy.ndarray) – Input samples.

• labels (numpy.ndarray) – Labels of input samples.

Returns

numpy.ndarray, loss of defense operation.

class mindarmour.adv_robustness.defenses.AdversarialDefenseWithAttacks(network, attacks, loss_fn=None, optimizer=None, bounds=(0.0, 1.0), replace_ratio=0.5)

Adversarial defense with attacks.

Parameters

• network (Cell) – A MindSpore network to be defensed.

• attacks (list[Attack]) – List of attack method.

• loss_fn (Functions) – Loss function. Default: None.

• optimizer (Cell) – Optimizer used to train the network. Default: None.

• bounds (tuple) – Upper and lower bounds of data. In form of (clip_min, clip_max). Default: (0.0, 1.0).

• replace_ratio (float) – Ratio of replacing original samples with adversarial, which must be between 0 and 1. Default: 0.5.

Raises

ValueError – If replace_ratio is not between 0 and 1.

Examples

>>> net = Net()
>>> fgsm = FastGradientSignMethod(net)
>>> pgd = ProjectedGradientDescent(net)
>>> ead = AdversarialDefenseWithAttacks(net, [fgsm, pgd])
>>> ead.defense(inputs, labels)

defense(inputs, labels)

Enhance model via training with adversarial examples generated from input samples.

Parameters

• inputs (numpy.ndarray) – Input samples.

• labels (numpy.ndarray) – Labels of input samples.

Returns

numpy.ndarray, loss of adversarial defense operation.

class mindarmour.adv_robustness.defenses.EnsembleAdversarialDefense(network, attacks, loss_fn=None, optimizer=None, bounds=(0.0, 1.0), replace_ratio=0.5)

Ensemble adversarial defense.

Parameters

• network (Cell) – A MindSpore network to be defensed.

• attacks (list[Attack]) – List of attack method.

• loss_fn (Functions) – Loss function. Default: None.

• optimizer (Cell) – Optimizer used to train the network. Default: None.

• bounds (tuple) – Upper and lower bounds of data. In form of (clip_min, clip_max). Default: (0.0, 1.0).

• replace_ratio (float) – Ratio of replacing original samples with adversarial, which must be between 0 and 1. Default: 0.5.

Raises

ValueError – If replace_ratio is not between 0 and 1.

Examples

>>> net = Net()
>>> fgsm = FastGradientSignMethod(net)
>>> pgd = ProjectedGradientDescent(net)
>>> ead = EnsembleAdversarialDefense(net, [fgsm, pgd])
>>> ead.defense(inputs, labels)

class mindarmour.adv_robustness.defenses.NaturalAdversarialDefense(network, loss_fn=None, optimizer=None, bounds=(0.0, 1.0), replace_ratio=0.5, eps=0.1)

Adversarial training based on FGSM.

Reference: A. Kurakin, et al., “Adversarial machine learning at scale,” in ICLR, 2017.

Parameters

• network (Cell) – A MindSpore network to be defensed.

• loss_fn (Functions) – Loss function. Default: None.

• optimizer (Cell) – Optimizer used to train the network. Default: None.

• bounds (tuple) – Upper and lower bounds of data. In form of (clip_min, clip_max). Default: (0.0, 1.0).

• replace_ratio (float) – Ratio of replacing original samples with adversarial samples. Default: 0.5.

• eps (float) – Step size of the attack method(FGSM). Default: 0.1.

Examples

>>> net = Net()
>>> adv_defense = NaturalAdversarialDefense(net)
>>> adv_defense.defense(inputs, labels)

class mindarmour.adv_robustness.defenses.ProjectedAdversarialDefense(network, loss_fn=None, optimizer=None, bounds=(0.0, 1.0), replace_ratio=0.5, eps=0.3, eps_iter=0.1, nb_iter=5, norm_level='inf')

Adversarial training based on PGD.

Reference: A. Madry, et al., “Towards deep learning models resistant to adversarial attacks,” in ICLR, 2018.

Parameters

• network (Cell) – A MindSpore network to be defensed.

• loss_fn (Functions) – Loss function. Default: None.

• optimizer (Cell) – Optimizer used to train the nerwork. Default: None.

• bounds (tuple) – Upper and lower bounds of input data. In form of (clip_min, clip_max). Default: (0.0, 1.0).

• replace_ratio (float) – Ratio of replacing original samples with adversarial samples. Default: 0.5.

• eps (float) – PGD attack parameters, epsilon. Default: 0.3.

• eps_iter (int) – PGD attack parameters, inner loop epsilon. Default:0.1.

• nb_iter (int) – PGD attack parameters, number of iteration. Default: 5.

• norm_level (str) – Norm type. ‘inf’ or ‘l2’. Default: ‘inf’.

Examples

>>> net = Net()
>>> adv_defense = ProjectedAdversarialDefense(net)
>>> adv_defense.defense(inputs, labels)