Read a tensor, i.e. numpy array, from HDF5. :param filepath: path to file to read :type filepath: str :param key: key to read :t

Read a JSON file. :param file: path to file to read :type file: str :return: parsed JSON as dict :rtype: dict

Read pickle file. :param file: path to file to read :type file: str :return: mixed :rtype: mixed

Computes the ROC curve for correct classified vs. incorrect classified. :return: false positive rates, true positive rates, threshol

Computes the ROC curve for correct classified vs. incorrect classified. :return: false positive rates, true positive rates, threshol

Computes the ROC curve for correct classified vs. incorrect classified. :return: false positive rates, true positive rates, threshol

Computes the ROC curve for correct classified vs. incorrect classified. :return: false positive rates, true positive rates, threshol

Remove a file. :param filepath: path to file :type filepath: str

Reset timer.

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

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

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

Robust test error curve. :return: robust test errors and thresholds :rtype: numpy.ndarray, numpy.ndarray

Run attack. :param model: model to attack :type model: torch.nn.Module :param images: images :type images: t

Run attack. :param model: model to attack :type model: torch.nn.Module :param images: images :type images: t

Run attack. :param model: model to attack :type model: torch.nn.Module :param images: images :type images: t

Run attack. :param model: model to attack :type model: torch.nn.Module :param images: images :type images: t

Run attack. :param model: model to attack :type model: torch.nn.Module :param images: images :type images: t

Run attack. :param model: model to attack :type model: torch.nn.Module :param images: images :type images: t

Run LaVAN attack :param model: model to attack, must contain normalization layer to work correctly :type model: torch.nn.Mod

Run attack. :param model: model to attack :type model: torch.nn.Module :param images: images :type images: t

Run attack. :param model: model to attack :type model: torch.nn.Module :param images: images :type images: t

Run attack. :param model: model to attack :type model: torch.nn.Module :param images: images :type images: t

Sample from Gaussian using antithetic sampling as described in [1]. [1] Andrew Ilyas, Logan Engstrom, Anish Athalye,

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

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

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

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

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

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

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

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

Success rate curve. :return: success rates and confidences :rtype: numpy.ndarray, numpy.ndarray

Helper for summaries, converts array of values to dict; :param values: values :type values: numpy.ndarray or torch.Tensor

Supplementary LaTeX table.

SVHN test images. :return: filepath :rtype: str

SVHN test labels. :return: filepath :rtype: str

SVHN train images. :return: filepath :rtype: str

SVHN train labels. :return: filepath :rtype: str

Get tensor or single value. :param mixed: variable, tensor or value :type mixed: mixed :return: tensor or value :rtype: torch.Te

Test a model on a clean or adversarial dataset. :param model: model :type model: torch.nn.Module :param testset: test set :type