Remove (Fashion-)MNIST

README.md

@@ -19,11 +19,6 @@ In particular, each module defines some specific task in the attack-defense chai
 
 ### Existing modules:
 1. `base_trainer`: Configured to poison and train a model on any of the supported datasets.
-1. `base_rep_saver`: Configured to extract representations from a model poisoned on any of the supported datasets.
-1. `base_grad_saver`: Configured to extract gradients from a model on poisoned on any of the supported datasets.
-1. `base_defense`: Configured to implement a defense based on the class representations on poisoned CIFAR-10. At the moment implements three defenses: PCA, k-means, and SPECTRE.
-1. `sever`: Configured to implement a defense based on the gradients of poisoned MNIST / Fashion-MNIST images. Referenced [here](#supported-defenses).
-1. `bgmd`: Configured to implement a defense based on an efficient implementation of the geometric median. Referenced [here](#supported-defenses).
 1. `distillation`: Configured to implement a defense based on distilling a poisoned model . Referenced [#TODO]().
 1. `base_utils`: Utility module, used by the base modules.
 
@@ -45,7 +40,6 @@ More documentation can be found in the `schemas` folder.
 ### Supported Datasets:
 1. Learning Multiple Layers of Features from Tiny Images [(Krizhevsky, 2009)](https://www.cs.toronto.edu/~kriz/learning-features-2009-TR.pdf).
 1. Gradient-based learning applied to document recognition [(LeCun et al., 1998)](http:https://yann.lecun.com/exdb/publis/pdf/lecun-98.pdf).
-1. Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms [(Xiao et al., 2017)](https://arxiv.org/pdf/1708.07747.pdf).
 
 ---
 ## Installation

modules/base_utils/datasets.py

@@ -69,42 +69,6 @@
 )
 
 
-MNIST_TRANSFORM_NORMALIZE_MEAN = (0.1307,)
-MNIST_TRANSFORM_NORMALIZE_STD = (0.3081,)
-MNIST_TRANSFORM_NORMALIZE = transforms.Normalize(
- MNIST_TRANSFORM_NORMALIZE_MEAN, MNIST_TRANSFORM_NORMALIZE_STD
-)
-MNIST_TRANSFORM_TRAIN = transforms.Compose(
- [
- transforms.ToTensor(),
- MNIST_TRANSFORM_NORMALIZE,
- ]
-)
-MNIST_TRANSFORM_TEST = transforms.Compose(
- [
- transforms.ToTensor(),
- MNIST_TRANSFORM_NORMALIZE,
- ]
-)
-
-FMNIST_TRANSFORM_NORMALIZE_MEAN = (0.2859,)
-FMNIST_TRANSFORM_NORMALIZE_STD = (0.3530,)
-FMNIST_TRANSFORM_NORMALIZE = transforms.Normalize(
- FMNIST_TRANSFORM_NORMALIZE_MEAN, FMNIST_TRANSFORM_NORMALIZE_STD
-)
-FMNIST_TRANSFORM_TRAIN = transforms.Compose(
- [
- transforms.ToTensor(),
- FMNIST_TRANSFORM_NORMALIZE,
- ]
-)
-FMNIST_TRANSFORM_TEST = transforms.Compose(
- [
- transforms.ToTensor(),
- FMNIST_TRANSFORM_NORMALIZE,
- ]
-)
-
 TINY_IMAGENET_TRANSFORM_NORMALIZE_MEAN = (0.485, 0.456, 0.406)
 TINY_IMAGENET_TRANSFORM_NORMALIZE_STD = (0.229, 0.224, 0.225)
 TINY_IMAGENET_TRANSFORM_NORMALIZE = transforms.Normalize(
@@ -128,32 +92,24 @@
 PATH = {
  'cifar': Path("./data/data_cifar10"),
  'cifar_100': Path("./data/data_cifar100"),
- 'mnist': Path("./data/data_mnist"),
- 'fmnist': Path("./data/data_fashion_mnist"),
  'tiny_imagenet': "/scr/tiny-imagenet-200"
 }
 
 TRANSFORM_TRAIN_XY = {
  'cifar': lambda xy: (CIFAR_TRANSFORM_TRAIN(xy[0]), xy[1]),
  'cifar_100': lambda xy: (CIFAR_100_TRANSFORM_TRAIN(xy[0]), xy[1]),
- 'mnist': lambda xy: (MNIST_TRANSFORM_TRAIN(xy[0]), xy[1]),
- 'fmnist': lambda xy: (FMNIST_TRANSFORM_TRAIN(xy[0]), xy[1]),
  'tiny_imagenet': lambda xy: (TINY_IMAGENET_TRANSFORM_TRAIN(xy[0]), xy[1])
 }
 
 TRANSFORM_TEST_XY = {
  'cifar': lambda xy: (CIFAR_TRANSFORM_TEST(xy[0]), xy[1]),
  'cifar_100': lambda xy: (CIFAR_100_TRANSFORM_TEST(xy[0]), xy[1]),
- 'mnist': lambda xy: (MNIST_TRANSFORM_TEST(xy[0]), xy[1]),
- 'fmnist': lambda xy: (FMNIST_TRANSFORM_TEST(xy[0]), xy[1]),
  'tiny_imagenet': lambda xy: (TINY_IMAGENET_TRANSFORM_TEST(xy[0]), xy[1])
 }
 
 N_CLASSES = {
  'cifar': 10,
  'cifar_100': 100,
- 'mnist': 10,
- 'fmnist': 10,
  'tiny_imagenet': 200
 }
 
@@ -436,30 +392,12 @@ def load_dataset(dataset_flag, train=True):
  return load_cifar_dataset(path, train)
  if dataset_flag == 'cifar_100':
  return load_cifar_100_dataset(path, train)
- elif dataset_flag == 'mnist':
- return load_mnist_dataset(path, train)
- elif dataset_flag == 'fmnist':
- return load_fmnist_dataset(path, train)
  elif dataset_flag == 'tiny_imagenet':
  return load_tiny_imagenet_dataset(path, train)
  else:
  raise NotImplementedError(f"Dataset {dataset_flag} is not supported.")
 
 
-def load_mnist_dataset(path, train=True):
- dataset = datasets.MNIST(root=str(path),
- train=train,
- download=True)
- return dataset
-
-
-def load_fmnist_dataset(path, train=True):
- dataset = datasets.FashionMNIST(root=str(path),
- train=train,
- download=True)
- return dataset
-
-
 def load_cifar_dataset(path, train=True):
  dataset = datasets.CIFAR10(root=str(path),
  train=train,
@@ -522,8 +460,6 @@ def make_dataloader(
 def pick_poisoner(poisoner_flag, dataset_flag, target_label):
  if dataset_flag == "cifar" or dataset_flag == "cifar_100":
  x_poisoner, all_x_poisoner = pick_cifar_poisoner(poisoner_flag)
- elif dataset_flag == "mnist" or dataset_flag == "fmnist":
- x_poisoner, all_x_poisoner = pick_mnist_poisoner(poisoner_flag)
  elif dataset_flag == "tiny_imagenet":
  x_poisoner, all_x_poisoner = pick_tiny_imagenet_poisoner(poisoner_flag)
  else:
@@ -669,58 +605,6 @@ def pick_tiny_imagenet_poisoner(poisoner_flag):
  return x_poisoner, all_x_poisoner
 
 
-def pick_mnist_poisoner(poisoner_flag):
- if poisoner_flag == "1xp":
- x_poisoner = PixelPoisoner(pos=(27, 27), col=(255))
- all_x_poisoner = PixelPoisoner(pos=(27, 27), col=(255))
- elif poisoner_flag == "3xp":
- x_poisoner = MultiPoisoner(
- [
- PixelPoisoner(pos=(27, 27), col=(255)),
- PixelPoisoner(pos=(27, 25), col=(255)),
- PixelPoisoner(pos=(25, 27), col=(255)),
- ]
- )
- all_x_poisoner = MultiPoisoner(
- [
- PixelPoisoner(pos=(27, 27), col=(255)),
- PixelPoisoner(pos=(27, 25), col=(255)),
- PixelPoisoner(pos=(25, 27), col=(255)),
- ]
- )
- elif poisoner_flag == "9xp":
- x_poisoner = MultiPoisoner(
- [
- PixelPoisoner(pos=(27, 27), col=(255)),
- PixelPoisoner(pos=(27, 25), col=(255)),
- PixelPoisoner(pos=(25, 27), col=(255)),
- PixelPoisoner(pos=(25, 25), col=(255)),
- PixelPoisoner(pos=(27, 23), col=(255)),
- PixelPoisoner(pos=(23, 27), col=(255)),
- PixelPoisoner(pos=(23, 23), col=(255)),
- PixelPoisoner(pos=(23, 25), col=(255)),
- PixelPoisoner(pos=(25, 23), col=(255)),
- ]
- )
- all_x_poisoner = MultiPoisoner(
- [
- PixelPoisoner(pos=(27, 27), col=(255)),
- PixelPoisoner(pos=(27, 25), col=(255)),
- PixelPoisoner(pos=(25, 27), col=(255)),
- PixelPoisoner(pos=(25, 25), col=(255)),
- PixelPoisoner(pos=(27, 23), col=(255)),
- PixelPoisoner(pos=(23, 27), col=(255)),
- PixelPoisoner(pos=(23, 23), col=(255)),
- PixelPoisoner(pos=(23, 25), col=(255)),
- PixelPoisoner(pos=(25, 23), col=(255)),
- ]
- )
- else:
- raise NotImplementedError()
-
- return x_poisoner, all_x_poisoner
-
-
 def generate_datasets(
  dataset_flag,
  poisoner,

schemas/base_trainer.toml

@@ -7,7 +7,7 @@
 [base_trainer]
 output = "string: Path to .pth file."
 model = "string: (r32p, r18, r18_tin, vgg, vgg_pretrain, vit_pretrain). For ResNets, VGG-19s, and ViTs"
-dataset = "string: (cifar / mnist / fmnist). For CIFAR-10 and MNIST / Fashion-MNIST datasets"
+dataset = "string: (cifar / cifar_100 / tiny_imagenet). For CIFAR-10, CIFAR-100 and Tiny Imagenet datasets"
 trainer = "string: (sgd / adam). Specifies optimizer. "
 source_label = "int: {0,1,...,9}. Specifies label to mimic"
 target_label = "int: {0,1,...,9}. Specifies label to attack"

schemas/distillation.toml

@@ -9,7 +9,7 @@
 output_path = "string: Path to .pth file."
 teacher_model = "string: (r32p, r18, r18_tin, vgg, vgg_pretrain, vit_pretrain). For ResNets, VGG-19s, and ViTs"
 student_model = "string: (r32p, r18, r18_tin, vgg, vgg_pretrain, vit_pretrain). For ResNets, VGG-19s, and ViTs"
-dataset = "string: (cifar / mnist / fmnist). For CIFAR-10 and MNIST / Fashion-MNIST datasets"
+dataset = "string: (cifar / cifar_100 / tiny_imagenet). For CIFAR-10, CIFAR-100 and Tiny Imagenet datasets"
 distill_percentage = "TODO"
 trainer = "string: (sgd / adam). Specifies optimizer. "
 source_label = "int: {0,1,...,9}. Specifies label to mimic"

schemas/downstream.toml

@@ -9,7 +9,7 @@
 input = "TODO"
 output_path = "string: Path to .pth file."
 downstream_model = "string: (r32p, r18, r18_tin, vgg, vgg_pretrain, vit_pretrain). For ResNets, VGG-19s, and ViTs"
-dataset = "string: (cifar / mnist / fmnist). For CIFAR-10 and MNIST / Fashion-MNIST datasets"
+dataset = "string: (cifar / cifar_100 / tiny_imagenet). For CIFAR-10, CIFAR-100 and Tiny Imagenet datasets"
 trainer = "string: (sgd / adam). Specifies optimizer. "
 source_label = "int: {0,1,...,9}. Specifies label to mimic"
 target_label = "int: {0,1,...,9}. Specifies label to attack"

schemas/mtt_labels.toml

@@ -11,7 +11,7 @@ opt_input = "TODO"
 output_path = "string: Path to .pth file."
 expert_model = "string: (r32p, r18, r18_tin, vgg, vgg_pretrain, vit_pretrain). For ResNets, VGG-19s, and ViTs"
 downstream_model = "string: (r32p, r18, r18_tin, vgg, vgg_pretrain, vit_pretrain). For ResNets, VGG-19s, and ViTs"
-dataset = "string: (cifar / mnist / fmnist). For CIFAR-10 and MNIST / Fashion-MNIST datasets"
+dataset = "string: (cifar / cifar_100 / tiny_imagenet). For CIFAR-10, CIFAR-100 and Tiny Imagenet datasets"
 trainer = "string: (sgd / adam). Specifies optimizer. "
 source_label = "int: {0,1,...,9}. Specifies label to mimic"
 target_label = "int: {0,1,...,9}. Specifies label to attack"