secmlt.manipulations package

Submodules

secmlt.manipulations.manipulation module

Manipulations for perturbing input samples.

class secmlt.manipulations.manipulation.AdditiveManipulation(domain_constraints: list[Constraint], perturbation_constraints: list[Constraint])

Bases: Manipulation

Additive manipulation for input data.

class secmlt.manipulations.manipulation.Manipulation(domain_constraints: list[Constraint], perturbation_constraints: list[Constraint])

Bases: ABC

Abstract class for manipulations.

__call__(x: torch.Tensor, delta: torch.Tensor) → tuple[torch.Tensor, torch.Tensor]

Apply the manipulation to the input data.

Parameters:

• x (torch.Tensor) – Input data.

• delta (torch.Tensor) – Perturbation to apply.

Returns:

Perturbed data and perturbation after the application of constraints.

Return type:

tuple[torch.Tensor, torch.Tensor]

__init__(domain_constraints: list[Constraint], perturbation_constraints: list[Constraint]) → None

Create manipulation object.

Parameters:

• domain_constraints (list[Constraint]) – Constraints for the domain bounds (x_adv).

• perturbation_constraints (list[Constraint]) – Constraints for the perturbation (delta).

abstract _apply_manipulation(x: torch.Tensor, delta: torch.Tensor) → torch.Tensor

Apply the manipulation.

Parameters:

• x (torch.Tensor) – Input samples.

• delta (torch.Tensor) – Manipulation to apply.

Returns:

Perturbed samples.

Return type:

torch.Tensor

property domain_constraints: list[Constraint]

Get the domain constraints for the manipulation.

Returns:

List of domain constraints for the manipulation.

Return type:

list[Constraint]

property perturbation_constraints: list[Constraint]

Get the perturbation constraints for the manipulation.

Returns:

List of perturbation constraints for the manipulation.

Return type:

list[Constraint]

Module contents

Functionalities for applying manipulations to input data.