Helper class for logging. Arguments: path (str): Path to log file.

Pre-activation Resnet model for TI-200 dataset.

Pre-activation Resnet model

Pre-activation Resnet model

ResNet model Arguments: block (BasicBlock or Bottleneck): type of basic block to be used. num_blocks (list): number of blocks

Implements a network block module for WideResnets. Arguments: nb_layers (int): number of layers. in_planes (int): number of i

PreActResNet model Arguments: num_classes (int): number of output classes. depth (int): number of layers. width (int)

Soft cross entropy loss with label smoothing.

Helper class for training a deep neural network. Arguments: info (dict): dataset information. args (dict): input arguments.

WideResNet model Arguments: num_classes (int): number of output classes. depth (int): number of layers. width (int):

WideResNet block group. Arguments: in_planes (int): number of input planes. out_planes (int): number of output filters.

A dataset with auxiliary pseudo-labeled data for CIFAR10.

A dataset with auxiliary pseudo-labeled data for CIFAR100.

Torch dataset for reading examples with corresponding perturbations. Arguments: root (str): path to saved data. transform (to

Cosine annealing LR schedule (used in Carmon et al, 2019).

One step fast gradient method. Perturbs the input with gradient (not gradient sign) of the loss wrt the input. Arguments: predict (nn

One step fast gradient sign method (Goodfellow et al, 2014). Arguments: predict (nn.Module): forward pass function. loss_fn (

APGD attack (from AutoAttack) with order=L2. The attack performs nb_iter steps of adaptive size, while always staying within eps from the ini

DeepFool Attack with order=L2. Arguments: predict (nn.Module): forward pass function. overshoot (float): how much to overshoo

PGD Attack with order=L2 Arguments: predict (nn.Module): forward pass function. loss_fn (nn.Module): loss function. e

APGD attack (from AutoAttack) with order=Linf. The attack performs nb_iter steps of adaptive size, while always staying within eps from the i

DeepFool Attack with order=Linf. Arguments: Arguments: predict (nn.Module): forward pass function. overshoot (float): how

PGD Attack with order=Linf Arguments: predict (nn.Module): forward pass function. loss_fn (nn.Module): loss function.

Standardizes the input data. Arguments: mean (list): mean. std (float): standard deviation. Returns: (input - mea

Transforms a numpy.ndarray to torch.Tensor.

Balanced sampling from the labeled and unlabeled data.

Helper class for training a deep neural network with model weight averaging (identical to Gowal et al, 2020). Arguments: info (dict):

WideResNet model Arguments: depth (int): number of layers. num_classes (int): number of output classes. widen_factor

WideResNet Block. Arguments: in_planes (int): number of input planes. out_planes (int): number of output filters. str

PreAct ResNet Block. Arguments: in_planes (int): number of input planes. out_planes (int): number of output filters.

APGD attack (from AutoAttack) (Croce et al, 2020). The attack performs nb_iter steps of adaptive size, while always staying within eps from t

Abstract base class for all attack classes. Arguments: predict (nn.Module): forward pass function. loss_fn (nn.Module): loss

Implements a basic block module for Resnets. Arguments: in_planes (int): number of input planes. out_planes (int): number of

Implements a basic block module for WideResNets. Arguments: in_planes (int): number of input planes. out_planes (int): number

Implements a basic block module with bottleneck for Resnets. Arguments: in_planes (int): number of input planes. out_planes (

DeepFool attack. [Seyed-Mohsen Moosavi-Dezfooli, Alhussein Fawzi, Pascal Frossard, "DeepFool: a simple and accurate method to fool deep

The projected gradient descent attack (Madry et al, 2017). The attack performs nb_iter steps of size eps_iter, while always staying within ep

Pre-activation version of the BasicBlock. Arguments: in_planes (int): number of input planes. planes (int): number of output

Pre-activation version of the BasicBlock for Resnets. Arguments: in_planes (int): number of input planes. planes (int): numbe

Pre-activation version of the BasicBlock for Resnets. Arguments: in_planes (int): number of input planes. planes (int): numbe

Pre-activation version of the original Bottleneck module. Arguments: in_planes (int): number of input planes. planes (int): n

Pre-activation version of the original Bottleneck module for Resnets. Arguments: in_planes (int): number of input planes. pla

Pre-activation version of the original Bottleneck module for Resnets. Arguments: in_planes (int): number of input planes. pla

A dataset with auxiliary pseudo-labeled data.