2. aijack.defense package

2.1. Subpackages

• 2.1.1. aijack.defense.crobustness package
  • 2.1.1.1. Submodules
  • 2.1.1.2. aijack.defense.crobustness.pixeldp module
  • 2.1.1.3. Module contents
• 2.1.2. aijack.defense.debugging package
  • 2.1.2.1. Subpackages
    • 2.1.2.1.1. aijack.defense.debugging.assertions package
      • 2.1.2.1.1.1. Submodules
      • 2.1.2.1.1.2. aijack.defense.debugging.assertions.assertions module
      • 2.1.2.1.1.3. aijack.defense.debugging.assertions.utils module
      • 2.1.2.1.1.4. Module contents
    • 2.1.2.1.2. aijack.defense.debugging.neuroncoverage package
      • 2.1.2.1.2.1. Submodules
      • 2.1.2.1.2.2. aijack.defense.debugging.neuroncoverage.neuroncoverage module
      • 2.1.2.1.2.3. Module contents
  • 2.1.2.2. Module contents
• 2.1.3. aijack.defense.dp package
  • 2.1.3.1. Subpackages
    • 2.1.3.1.1. aijack.defense.dp.manager package
      • 2.1.3.1.1.1. Submodules
      • 2.1.3.1.1.2. aijack.defense.dp.manager.accountant module
      • 2.1.3.1.1.3. aijack.defense.dp.manager.adadps module
      • 2.1.3.1.1.4. aijack.defense.dp.manager.client module
      • 2.1.3.1.1.5. aijack.defense.dp.manager.dataloader module
      • 2.1.3.1.1.6. aijack.defense.dp.manager.dp_manager module
      • 2.1.3.1.1.7. aijack.defense.dp.manager.dpoptimizer module
      • 2.1.3.1.1.8. aijack.defense.dp.manager.rdp module
      • 2.1.3.1.1.9. aijack.defense.dp.manager.utils module
      • 2.1.3.1.1.10. Module contents
  • 2.1.3.2. Module contents
• 2.1.4. aijack.defense.foolsgold package
  • 2.1.4.1. Submodules
  • 2.1.4.2. aijack.defense.foolsgold.server module
  • 2.1.4.3. Module contents
• 2.1.5. aijack.defense.kanonymity package
  • 2.1.5.1. Submodules
  • 2.1.5.2. aijack.defense.kanonymity.wrapper module
  • 2.1.5.3. Module contents
• 2.1.6. aijack.defense.mid package
  • 2.1.6.1. Submodules
  • 2.1.6.2. aijack.defense.mid.loss module
  • 2.1.6.3. aijack.defense.mid.nn module
  • 2.1.6.4. Module contents
• 2.1.7. aijack.defense.paillier package
  • 2.1.7.1. Submodules
  • 2.1.7.2. aijack.defense.paillier.fed_wrapper module
  • 2.1.7.3. aijack.defense.paillier.torch_wrapper module
  • 2.1.7.4. Module contents
• 2.1.8. aijack.defense.soteria package
  • 2.1.8.1. Submodules
  • 2.1.8.2. aijack.defense.soteria.soteria_client module
  • 2.1.8.3. Module contents
• 2.1.9. aijack.defense.sparse package
  • 2.1.9.1. Submodules
  • 2.1.9.2. aijack.defense.sparse.topk module
  • 2.1.9.3. Module contents

2.2. Submodules

2.3. aijack.defense.purifier module

aijack.defense.purifier.PurifierLoss(prediction, pred_purified, lam=0.2, purifier_criterion=MSELoss(), accuracy_criterion=CrossEntropyLoss())

basic loss function for purification

reference https://arxiv.org/abs/2005.03915

train purifier G against target model F to minimize the following objective function

L(G) = E[R(G(F(x)), F(x)) + λC(G(F(x), argmax F(x)))]

R is a reconstruction loss function C is a cross entropy loss function λ controls the balance of the two loss functions

Parameters

• prediction – predicted value of target model

• pred_purified – purified predicted value of target model

• lam – controls the balance of the following two functions

• purifier_criterion – loss function to reshapre confidense score

• accuracy_criterion – loss function to preserve the accuracy (C)

Returns

weighted average of the two loss function

Return type

loss_purifier

class aijack.defense.purifier.Purifier_Cifar10

Bases: torch.nn.modules.module.Module

autoencoder for purification on Cifar10 reference https://arxiv.org/abs/2005.03915

forward(x)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

2.4. Module contents

Subpackage for defense algorithms for machine learning models.