Normal training configuration. :param directory: directory for models :type directory: str

L_inf normalized PGD implementation. :param base_lr: learning rate :type base_lr: float :param max_iterations: iterations :type

Add scalar values. :param tag: tag for scalar :type tag: str :param tag_scalar_dict: values :type tag_scalar

Forward pass. :param input: input :type input: torch.autograd.Variable :return: output :rtype: torch.autogra

Main.

Report progress. :param epoch: epoch :type epoch: int :param batch: batch :type batch: int :param batches: batches :type

Computes the ROC curve for correct classified vs. incorrect classified. :return: ROC AUC score :rtype: float

Labels and scores for ROC. :return: labels and scores for sklearn.metrics.roc_auc_score :rtype: numpy.ndarray, numpy.ndarray

Labels and scores for ROC. :return: returns labels and scores for sklearn.metrics.roc_auc_score :rtype: numpy.ndarray, numpy

Define labels and scores for ROC. :return: labels and scores for sklearn.metrics.roc_auc_score :rtype: numpy.ndarray, numpy.

Labels and scores for ROC. :return: labels and scores for sklearn.metrics.roc_auc_score :rtype: numpy.ndarray, numpy.ndarray

Test error at 95%TPR. :return: robust test error :rtype: float

CCAT configuration. :param directory: directory for models :type directory: str :param attack: attack for adversarial training :

Adversarial frames attack. :param base_lr: learning rate :type base_lr: float :param max_iterations: iterations :type max_iterat

L_inf geometry attack. :param max_iterations: iterations :type max_iterations: int :param epsilon: epsilon :type epsilon: float

Get augmentation using imgaug. :param noise: whether to add additive noise :type noise: bool :param crop: whether to apply cropping

Constructor. Exactly one of include_list and exclude_list must be specified. :param img_dims: image dimensions :type img_dim

L_0 geometry attack. :param max_iterations: iterations :type max_iterations: int :param epsilon: epsilon :type epsilon: float

L_0 random sampling. :param max_iterations: iterations :type max_iterations: int :param epsilon: epsilon :type epsilon: float

L_1 geometry attack. :param max_iterations: iterations :type max_iterations: int :param epsilon: epsilon :type epsilon: float

L_1 random sampling. :param max_iterations: iterations :type max_iterations: int :param epsilon: epsilon :type epsilon: float

L_2 geometry attack. :param max_iterations: iterations :type max_iterations: int :param epsilon: epsilon :type epsilon: float

L_2 random sampling. :param max_iterations: iterations :type max_iterations: int :param epsilon: epsilon :type epsilon: float

L_inf normalized PGD with random and zero initialization for training. :param base_lr: learning rate :type base_lr: float :param max

L_inf random sampling. :param max_iterations: iterations :type max_iterations: int :param epsilon: epsilon :type epsilon: float

L_inf reference PGD attack. :param base_lr: learning rate :type base_lr: float :param max_iterations: iterations :type max_itera

L_inf simple black-box attack. :param max_iterations: iterations :type max_iterations: int :param epsilon: epsilon for attack :t

Distal or OOD supplementary LaTeX table.

Add figure. :param tag: tag :type tag: str :param figure: test :type figure: matplotlib.pyplot.figure

Add image. :param tag: tag :type tag: str :param img_tensor: image :type img_tensor: torch.Tensor or numpy.n

Add attack targets. :param targets: targets :type targets: numpy.ndarray

Attack model. :param model: model :type model: torch.nn.Module :param testset: test set :type testset: torch.utils.data.Data

Loss. :param logits: predicted logits :type logits: torch.autograd.Variable :param targets: target distributions :type targets:

Check h5py installation.

Checks packages that can be installed via PIP.

Check torch installation.

Convert MNIST.

Convert SVHN.

Simply write log message.

Download and convert Cifar10.

Evaluate.

Evaluate.

Evaluate.

Expands the tensor using view to allow broadcasting. :param array: input tensor :type array: numpy.ndarray :param array_as: referenc

State file. :param model_file: base state file :type model_file: str :return: state file of ongoing training :rtype: str

State file. :param model_file: base state file :type model_file: str :return: state file of ongoing training :rtype: str

State file. :param model_file: base state file :type model_file: str :return: state file of ongoing training :rtype: str

State file. :param model_file: base state file :type model_file: str :return: state file of ongoing training :rtype: str

Save.

Forward pass. :param input: input :type input: torch.autograd.Variable :return: output :rtype: torch.autogra

Get attacks to test.

Gets mask in image shape given mask coordinates :param mask_coords: mask coordinates for the batch of masks :type mask_coord

Get parser. :return: parser :rtype: argparse.ArgumentParser

Get parser. :return: parser :rtype: argparse.ArgumentParser

Get parser. :return: parser :rtype: argparse.ArgumentParser

Get parser. :return: parser :rtype: argparse.ArgumentParser

Get parser. :return: parser :rtype: argparse.ArgumentParser

Get parser. :return: parser :rtype: argparse.ArgumentParser

Get parser. :return: parser :rtype: argparse.ArgumentParser

Add scalar value. :param tag: tag for scalar :type tag: str :param value: value :type value: mixed :

Get attacks to test.

Get attacks to test.

Get attack for training.

Get attack for training.

Add a layer. :param index: index :type index: int :param name: layer name :type name: str :param lay

Load a model. :param model_file: model file :type model_file: str :param cuda: use on GPU :type cuda: bool :return: model

Main.

Main.

Main.

Main.

Main.

Main.

Get memory usage. :return: memory usage :rtype: str

Normalization. :param gradients: gradients :type gradients: torch.autograd.Variable

creates n_iter images which differ from orig_x in at most k pixels

returns a batch with the possible perturbations of the pixel in position pos

returns a batch with all the possible perturbations of the image orig_x

Close all open HDF5 files and report number of closed files. :return: number of closed files :rtype: int

Generates n mask coordinates randomly from allowed locations :param n: number of masks to generate, defaults to 1 :type n: i

Normalization. :param gradients: gradients :type gradients: torch.autograd.Variable

Setup.

Set dataloaders.

Set dataloaders.

Set dataloaders.

creates the sigma-map for the batch x