ART-Adversarial Robustness Toolbox检测AI模型及对抗攻击的工具

一、工具简介

Adversarial Robustness Toolbox 是 IBM 研究团队开源的用于检测模型及对抗攻击的工具箱，为开发人员加强 AI模型被误导的防御性，让 AI 系统变得更加安全，ART支持所有流行的机器学习框架（TensorFlow，Keras，PyTorch，MXNet，scikit-learn，XGBoost，LightGBM，CatBoost，GPy等），所有数据类型（图像，表格，音频，视频等）和机器学习任务（分类，物体检测，语音识别，生成模型，认证等）。

支持以下攻击方法:

Deep Fool
Fast Gradient Method
Jacobian Saliency Map
Universal Perturbation
Virtual Adversarial Method
C&W Attack
NewtonFool

支持以下防御方法:

Feature squeezing
Spatial smoothing
Label smoothing
Adversarial training
Virtual adversarial training

github地址：https://github.com/Trusted-AI/adversarial-robustness-toolbox

Get Started	Documentation	Contributing
- Installation - Examples - Notebooks	- Attacks - Defences - Estimators - Metrics - Technical Documentation	- Slack, Invitation - Contributing - Roadmap - Citing

二、实际应用-攻击样本生成

2.1、手写数字

攻击脚本：

import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
import numpy as np
import matplotlib.pyplot as plt

from art.attacks.evasion import FastGradientMethod
from art.estimators.classification import PyTorchClassifier
from art.utils import load_mnist
import warnings

warnings.filterwarnings("ignore")


class Net(nn.Module):
    """
    定义初始模型
    """

    def __init__(self):
        super(Net, self).__init__()
        self.conv_1 = nn.Conv2d(in_channels=1, out_channels=4, kernel_size=5, stride=1)
        self.conv_2 = nn.Conv2d(in_channels=4, out_channels=10, kernel_size=5, stride=1)
        self.fc_1 = nn.Linear(in_features=4 * 4 * 10, out_features=100)
        self.fc_2 = nn.Linear(in_features=100, out_features=10)

    def forward(self, x):
        x = F.relu(self.conv_1(x))
        x = F.max_pool2d(x, 2, 2)
        x = F.relu(self.conv_2(x))
        x = F.max_pool2d(x, 2, 2)
        x = x.view(-1, 4 * 4 * 10)
        x = F.relu(self.fc_1(x))
        x = self.fc_2(x)
        return x


if __name__ == '__main__':
    # 导入ART自带的MNIST数据集
    (x_train, y_train), (x_test, y_test), min_pixel_value, max_pixel_value = load_mnist()
    x_train = np.swapaxes(x_train, 1, 3).astype(np.float32)
    x_test = np.swapaxes(x_test, 1, 3).astype(np.float32)

    # 创建模型
    model = Net()
    criterion = nn.CrossEntropyLoss()
    optimizer = optim.Adam(model.parameters(), lr=0.01)

    # 创建并训练ART分类器,注意：框架内置的调用方式训练模型
    classifier = PyTorchClassifier(
        model=model,
        clip_values=(min_pixel_value, max_pixel_value),
        loss=criterion,
        optimizer=optimizer,
        input_shape=(1, 28, 28),
        nb_classes=10,
    )

    # 演示代码用训练集的前6000条样本训练模型
    classifier.fit(x_train[:6000], y_train[:6000], batch_size=128, nb_epochs=2)
    predictions = classifier.predict(x_test)
    accuracy = np.sum(np.argmax(predictions, axis=1) == np.argmax(y_test, axis=1)) / len(y_test)
    print("在原始测试集上的准确率为: {}%".format(accuracy * 100))

    # 用FGSM算法对测试集生成对抗样本并测试分类器对对抗样本的评估效果
    n = 1
    for e in range(5, 25, 5):
        print('-' * 88)
        eps = round(e / 100, 2)
        attack = FastGradientMethod(estimator=classifier, eps=eps)
        x_test_adv = attack.generate(x=x_test[:1000])
        predictions = classifier.predict(x_test_adv)
        accuracy = np.sum(np.argmax(predictions, axis=1) == np.argmax(y_test[:1000], axis=1)) / len(y_test[:1000])
        print(f"扰动eps={eps}时分类器的准确率为: {round(accuracy, 4) * 100}%")
        adv_data = np.squeeze(x_test_adv)
        plt.subplot(2, 2, n)
        plt.title(f'eps:{eps}')
        plt.imshow(adv_data[1])
        print("分类器将其分为：", np.argmax(classifier.predict(x_test_adv[1:2]), axis=1)[0])
        n += 1
    plt.show()

运行结果：

在原始测试集上的准确率为: 93.73%
----------------------------------------------------------------------------------------
扰动eps=0.05时分类器的准确率为: 81.6%
分类器将其分为： 2
----------------------------------------------------------------------------------------
扰动eps=0.1时分类器的准确率为: 57.8%
分类器将其分为： 2
----------------------------------------------------------------------------------------
扰动eps=0.15时分类器的准确率为: 34.5%
分类器将其分为： 2
----------------------------------------------------------------------------------------
扰动eps=0.2时分类器的准确率为: 16.400000000000002%
分类器将其分为： 0

分析结果：

根据以上结果，随着eps的增大，分类器的准确率在下降，当准确下降至16.4%时，分类器将‘2’预测为‘0’。