-
Notifications
You must be signed in to change notification settings - Fork 9
/
projected_gradient.py
204 lines (154 loc) · 8.96 KB
/
projected_gradient.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
import torch
import torch.nn as nn
from data import str2dataset
from model import str2model
from utils import *
from wasserstein_attack import WassersteinAttack
from projection import dual_projection, dual_capacity_constrained_projection
class ProjectedGradient(WassersteinAttack):
def __init__(self,
predict, loss_fn,
eps, kernel_size,
lr, nb_iter, dual_max_iter, grad_tol, int_tol,
device="cuda",
postprocess=False,
verbose=True,
):
super().__init__(predict=predict, loss_fn=loss_fn,
eps=eps, kernel_size=kernel_size,
device=device,
postprocess=postprocess,
verbose=verbose,
)
self.lr = lr
self.nb_iter = nb_iter
self.dual_max_iter = dual_max_iter
self.grad_tol = grad_tol
self.int_tol = int_tol
self.inf = 1000000
# self.capacity_proj_mod = capacity_proj_mod
def perturb(self, X, y):
batch_size, c, h, w = X.size()
self.initialize_cost(X, inf=self.inf)
pi = self.initialize_coupling(X).clone().detach().requires_grad_(True)
normalization = X.sum(dim=(1, 2, 3), keepdim=True)
for t in range(self.nb_iter):
adv_example = self.coupling2adversarial(pi, X)
scores = self.predict(adv_example.clamp(min=self.clip_min, max=self.clip_max))
loss = self.loss_fn(scores, y)
loss.backward()
with torch.no_grad():
self.lst_loss.append(loss.item())
self.lst_acc.append((scores.max(dim=1)[1] == y).sum().item())
"""Add a small constant to enhance numerical stability"""
# print(tensor_norm(pi.grad, p='inf').min())
pi.grad /= (tensor_norm(pi.grad, p='inf').view(batch_size, 1, 1, 1) + 1e-35)
assert (pi.grad == pi.grad).all() and (pi.grad != float('inf')).all() and (pi.grad != float('-inf')).all()
pi += self.lr * pi.grad
start = torch.cuda.Event(enable_timing=True)
end = torch.cuda.Event(enable_timing=True)
start.record()
# if self.capacity_proj_mod == -1:
pi, num_iter = dual_projection(pi,
X,
cost=self.cost,
eps=self.eps * normalization.squeeze(),
dual_max_iter=self.dual_max_iter,
grad_tol=self.grad_tol,
int_tol=self.int_tol)
# elif (t + 1) % self.capacity_proj_mod == 0:
# pi = dual_capacity_constrained_projection(pi,
# X,
# self.cost,
# eps=self.eps * normalization.squeeze(),
# transpose_idx=self.forward_idx,
# detranspose_idx=self.backward_idx,
# coupling2adversarial=self.coupling2adversarial)
# num_iter = 3000
end.record()
torch.cuda.synchronize()
self.run_time += start.elapsed_time(end)
self.num_iter += num_iter
self.func_calls += 1
if self.verbose and (t + 1) % 10 == 0:
print("num of iters : {:4d}, ".format(t + 1),
"loss : {:9.3f}, ".format(loss.item()),
"acc : {:5.2f}%, ".format((scores.max(dim=1)[1] == y).sum().item() / batch_size * 100),
"dual iter : {:4d}, ".format(num_iter),
"per iter time : {:7.3f}ms".format(start.elapsed_time(end) / num_iter))
self.check_nonnegativity(pi / normalization, tol=1e-6, verbose=False)
self.check_marginal_constraint(pi / normalization, X / normalization, tol=1e-6, verbose=False)
self.check_transport_cost(pi / normalization, tol=1e-3, verbose=False)
pi = pi.clone().detach().requires_grad_(True)
with torch.no_grad():
adv_example = self.coupling2adversarial(pi, X)
check_hypercube(adv_example, verbose=self.verbose)
self.check_nonnegativity(pi / normalization, tol=1e-5, verbose=self.verbose)
self.check_marginal_constraint(pi / normalization, X / normalization, tol=1e-5, verbose=self.verbose)
self.check_transport_cost(pi / normalization, tol=self.eps * 1e-3, verbose=self.verbose)
if self.postprocess is True:
if self.verbose:
print("==========> post-processing projection........")
pi = dual_capacity_constrained_projection(pi,
X,
self.cost,
eps=self.eps * normalization.squeeze(),
transpose_idx=self.forward_idx,
detranspose_idx=self.backward_idx,
coupling2adversarial=self.coupling2adversarial)
adv_example = self.coupling2adversarial(pi, X)
check_hypercube(adv_example, tol=self.eps * 1e-1, verbose=self.verbose)
self.check_nonnegativity(pi / normalization, tol=1e-6, verbose=self.verbose)
self.check_marginal_constraint(pi / normalization, X / normalization, tol=1e-6, verbose=self.verbose)
self.check_transport_cost(pi / normalization, tol=self.eps * 1e-3, verbose=self.verbose)
"""Do not clip the adversarial examples to preserve pixel mass"""
return adv_example
if __name__ == "__main__":
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('--dataset', type=str, default='MNIST')
parser.add_argument('--checkpoint', type=str, default='mnist_vanilla')
parser.add_argument('--batch_size', type=int, default=20)
parser.add_argument('--num_batch', type=int_or_none, default=5)
parser.add_argument('--eps', type=float, default=0.5, help='the perturbation size')
parser.add_argument('--kernel_size', type=int_or_none, default=5)
parser.add_argument('--lr', type=float, default=0.1, help='gradient step size')
parser.add_argument('--nb_iter', type=int, default=20)
parser.add_argument('--dual_max_iter', type=int, default=15)
parser.add_argument('--grad_tol', type=float_or_none, default=1e-4)
parser.add_argument('--int_tol', type=float_or_none, default=1e-4)
parser.add_argument('--save_img_loc', type=str_or_none, default=None)
parser.add_argument('--save_info_loc', type=str_or_none, default=None)
parser.add_argument('--seed', type=int, default=0)
parser.add_argument('--postprocess', type=str2bool, default=False)
# parser.add_argument('--capacity_proj_mod', type=int, default=-1)
args = parser.parse_args()
print(args)
device = "cuda"
set_seed(args.seed)
testset, normalize, unnormalize = str2dataset(args.dataset)
testloader = torch.utils.data.DataLoader(testset, batch_size=args.batch_size, shuffle=False, num_workers=0)
net = str2model(args.checkpoint, dataset=args.dataset, pretrained=True).eval().to(device)
for param in net.parameters():
param.requires_grad = False
projected_gradient = ProjectedGradient(predict=lambda x: net(normalize(x)),
loss_fn=nn.CrossEntropyLoss(reduction="sum"),
eps=args.eps,
kernel_size=args.kernel_size,
lr=args.lr,
nb_iter=args.nb_iter,
dual_max_iter=args.dual_max_iter,
grad_tol=args.grad_tol,
int_tol=args.int_tol,
device=device,
postprocess=args.postprocess,
verbose=True)
acc = test(lambda x: net(normalize(x)),
testloader,
device=device,
attacker=projected_gradient,
num_batch=args.num_batch,
save_img_loc=args.save_img_loc)
projected_gradient.print_info(acc)
if args.save_info_loc is not None:
projected_gradient.save_info(acc, args.save_info_loc)