def train_four_layer_mnist_victim(
transform=None,
batch_size=100,
num_workers=4,
rand_split_val=None,
gpus=None,
max_epochs=8,
learning_rate=1e-3,
):
"""Trains a 4-layer fully connected neural network on MNIST data
Parameters:
transform: A Torchvision.transforms transformation to be applied to MNIST data
batch_size: An integer of the size of batches to be trained and tested upon
num_workers: An integer number of workers assigned to computations
rand_split_val: An array describing how the val and train data are split
gpus: An integer num of gpus available to train upon
max_epochs: An integer of the maximum # of epochs to run
learning_rate: A float that is the learning rate for the optimizer
Returns:
Trained model ready for inference"""
input_size = 784 # 28*28 or the size of a single image
targets = 10 # the number of digits any image can possibly represent
# Uses all available GPUs for computation by default
if gpus is None:
gpus = torch.cuda.device_count()
if transform is None:
transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307, ), (0.3081, ))
])
if rand_split_val is None:
rand_split_val = [55000, 5000]
# Establish hyperparameters and DataLoaders
hparams = set_hparams(
transform,
batch_size,
num_workers,
gpus,
max_epochs,
learning_rate,
input_size,
targets,
)
train_dataloader, val_dataloader, test_dataloader = get_mnist_loaders(
hparams)
# Train, test, and convert the model to inference
mnist_model = train_and_test(FourLayerClassifier, train_dataloader,
val_dataloader, test_dataloader, hparams)
mnist_model = convert_to_inference(mnist_model)
return mnist_model
def get_substitute_model(self):
model = train_and_test(
self.substitute_model_arch,
self.train_dataloader,
self.valid_dataloader,
self.test_dataloader,
self.hparams,
)
model = convert_to_inference(model)
return model
def train_mnist_victim(
transform=None,
batch_size=100,
num_workers=4,
rand_split_val=None,
gpus=1,
max_epochs=8,
learning_rate=1e-3,
):
"""Trains a 3-layer fully connected neural network on MNIST data
Parameters:
transform: A Torchvision.transforms transformation to be applied to MNIST data
batch_size: An integer of the size of batches to be trained and tested upon
num_workers: An integer number of workers assigned to computations
rand_split_val: An array describing how the val and train data are split
gpus: An integer num of gpus available to train upon
max_epochs: An integer of the maximum # of epochs to run
learning_rate: A float that is the learning rate for the optimizer
Returns:
Trained model ready for inference"""
# Define hyperparameters implied by the use of MNIST
input_size = 784
targets = 10
if transform is None:
transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.1307, ), (0.3081, ))
])
if rand_split_val is None:
rand_split_val = [55000, 5000]
# Establish hyperparameters and DataLoaders
hparams = set_hparams(
transform,
batch_size,
num_workers,
# rand_split_val,
gpus,
max_epochs,
learning_rate,
input_size,
targets,
)
train_dataloader, val_dataloader, test_dataloader = get_mnist_loaders(
hparams)
# Train, test, and convert the model to inference
mnist_model = train_and_test(ThreeLayerClassifier, train_dataloader,
val_dataloader, test_dataloader, hparams)
mnist_model = convert_to_inference(mnist_model)
return mnist_model
def get_substitute_model(self):
print("Training the substitute_model")
model = train_and_test(
self.substitute_model_arch,
self.train_dataloader,
self.valid_dataloader,
self.test_dataloader,
self.hparams,
self.callback,
self.trainer_args,
)
# This may limit the attack to PyTorch Lightning substitutes
model = convert_to_inference(model)
return model