Download the file lenet.pt, which contains a LeNet model trained
on MNIST data with pixel values normalized to the range [0, 1] by dividing
them by 255.
The model can be loaded as follows.
import torch
import torch.nn as nn
import torch.nn.functional as F
class LeNet(nn.Module):
def __init__(self, nchannels, width, height):
super(LeNet, self).__init__()
self.conv1 = nn.Conv2d(nchannels, 6, 5)
self.conv2 = nn.Conv2d(6, 16, 5)
self.fc1 = nn.Linear(16*int((width-12)/4)*int((height-12)/4), 120)
self.fc2 = nn.Linear(120, 84)
self.fc3 = nn.Linear(84, 10)
def forward(self, X):
out = F.relu(self.conv1(X))
out = F.max_pool2d(out, 2)
out = F.relu(self.conv2(out))
out = F.max_pool2d(out, 2)
out = out.view(out.size(0), -1)
out = F.relu(self.fc1(out))
out = F.relu(self.fc2(out))
out = self.fc3(out)
return out
net = torch.load('lenet.pt')
Implement the fast gradient sign method, and use it to generate adversarial examples for the first 16 test images.