advsecurenet.shared.types.configs.attack_configs package

advsecurenet.shared.types.configs.attack_configs.attack_config module

class advsecurenet.shared.types.configs.attack_configs.attack_config.AttackConfig(*, device: ~advsecurenet.shared.types.configs.device_config.DeviceConfig = <factory>, targeted: bool = False)

Bases: object

This dataclass is used to store the base configuration of the attacks. It contains the device configuration and a flag to indicate if the attack is targeted or not.

device: DeviceConfig

targeted: bool = False

advsecurenet.shared.types.configs.attack_configs.attacker_config module

class advsecurenet.shared.types.configs.attack_configs.attacker_config.AttackerConfig(model: ~torch.nn.modules.module.Module, attack: ~advsecurenet.attacks.base.adversarial_attack.AdversarialAttack, dataloader: ~torch.utils.data.dataloader.DataLoader | ~advsecurenet.shared.types.configs.dataloader_config.DataLoaderConfig, device: ~advsecurenet.shared.types.configs.device_config.DeviceConfig, return_adversarial_images: bool | None = False, evaluators: list[str] | None = <factory>)

Bases: object

Configuration class for the Attacker module.

attack: AdversarialAttack

dataloader: DataLoader | DataLoaderConfig

device: DeviceConfig

evaluators: list[str] | None

model: Module

return_adversarial_images: bool | None = False

advsecurenet.shared.types.configs.attack_configs.cw_attack_config module

class advsecurenet.shared.types.configs.attack_configs.cw_attack_config.CWAttackConfig(*, device: ~advsecurenet.shared.types.configs.device_config.DeviceConfig = <factory>, targeted: bool = False, c_init: float = 0.1, kappa: float = 0, learning_rate: float = 0.01, max_iterations: int = 10, abort_early: bool = False, binary_search_steps: int = 10, clip_min: float = 0, clip_max: float = 1, c_lower: float = 1e-06, c_upper: float = 1, patience: int = 5)

Bases: AttackConfig

Carlini-Wagner attack configuration.

abort_early: bool = False

binary_search_steps: int = 10

c_init: float = 0.1

c_lower: float = 1e-06

c_upper: float = 1

clip_max: float = 1

clip_min: float = 0

kappa: float = 0

learning_rate: float = 0.01

max_iterations: int = 10

patience: int = 5

advsecurenet.shared.types.configs.attack_configs.decision_boundary_attack_config module

class advsecurenet.shared.types.configs.attack_configs.decision_boundary_attack_config.DecisionBoundaryAttackConfig(*, device: advsecurenet.shared.types.configs.device_config.DeviceConfig = <factory>, targeted: bool = False, initial_delta: float = 0.1, initial_epsilon: float = 0.1, max_delta_trials: int = 20, max_epsilon_trials: int = 20, max_iterations: int = 1000, max_initialization_trials: int = 100, step_adapt: float = 0.9, verbose: bool = False, early_stopping: bool = True, early_stopping_threshold: float = 0.0001, early_stopping_patience: int = 10)

Bases: AttackConfig

early_stopping: bool = True

early_stopping_patience: int = 10

early_stopping_threshold: float = 0.0001

initial_delta: float = 0.1

initial_epsilon: float = 0.1

max_delta_trials: int = 20

max_epsilon_trials: int = 20

max_initialization_trials: int = 100

max_iterations: int = 1000

step_adapt: float = 0.9

targeted: bool = False

verbose: bool = False

advsecurenet.shared.types.configs.attack_configs.deepfool_attack_config module

class advsecurenet.shared.types.configs.attack_configs.deepfool_attack_config.DeepFoolAttackConfig(*, device: ~advsecurenet.shared.types.configs.device_config.DeviceConfig = <factory>, targeted: bool = False, num_classes: int = 10, overshoot: float = 0.02, max_iterations: int = 50)

Bases: AttackConfig

DeepFool attack configuration.

max_iterations: int = 50

num_classes: int = 10

overshoot: float = 0.02

advsecurenet.shared.types.configs.attack_configs.fgsm_attack_config module

class advsecurenet.shared.types.configs.attack_configs.fgsm_attack_config.FgsmAttackConfig(*, device: ~advsecurenet.shared.types.configs.device_config.DeviceConfig = <factory>, targeted: bool = False, epsilon: float = 0.3)

Bases: AttackConfig

FGSM attack configuration.

epsilon: float = 0.3

advsecurenet.shared.types.configs.attack_configs.lots_attack_config module

class advsecurenet.shared.types.configs.attack_configs.lots_attack_config.LotsAttackConfig(*, device: ~advsecurenet.shared.types.configs.device_config.DeviceConfig = <factory>, targeted: bool = False, deep_feature_layer: str, mode: ~advsecurenet.shared.types.configs.attack_configs.lots_attack_config.LotsAttackMode = LotsAttackMode.SINGLE, epsilon: float = 0.1, learning_rate: float = 0.00392156862745098, max_iterations: int = 1000, verbose: bool = True)

Bases: AttackConfig

Configuration class for LotsAttack.

deep_feature_layer

The deep feature layer to be used.

Type:: str

mode

The mode of the LotsAttack. Defaults to LotsAttackMode.SINGLE.

Type:: LotsAttackMode

epsilon

The epsilon value for the attack. Defaults to 0.1.

Type:: float

learning_rate

The learning rate for the attack. Defaults to 1./255.

Type:: float

max_iterations

The maximum number of iterations for the attack. Defaults to 1000.

Type:: int

verbose

Whether to print verbose output during the attack. Defaults to True.

Type:: bool

device

The device to be used for the attack.

Type:: torch.device

deep_feature_layer: str

epsilon: float = 0.1

learning_rate: float = 0.00392156862745098

max_iterations: int = 1000

mode: LotsAttackMode = 'single'

verbose: bool = True

class advsecurenet.shared.types.configs.attack_configs.lots_attack_config.LotsAttackMode(value)

Bases: Enum

Enum class for the mode of the LotsAttack. Possible values are:

ITERATIVE: Iterative mode. Runs the attack for the specified number of iterations.
SINGLE: Single mode. Runs the attack for a single iteration.

ITERATIVE = 'iterative'

SINGLE = 'single'

advsecurenet.shared.types.configs.attack_configs.pgd_attack_config module

class advsecurenet.shared.types.configs.attack_configs.pgd_attack_config.PgdAttackConfig(*, device: ~advsecurenet.shared.types.configs.device_config.DeviceConfig = <factory>, targeted: bool = False, epsilon: float = 0.3, alpha: float = 0.00784313725490196, num_iter: int = 40)

Bases: AttackConfig

PGD attack configuration.

alpha: float = 0.00784313725490196

epsilon: float = 0.3

num_iter: int = 40