MindArmour 使用

举报
irrational 发表于 2022/01/17 23:01:46 2022/01/17
【摘要】 文章首发于笔者华为云,欢迎关注MindArmour 使用(万字详解,gitee同步发表)_MindSpore_昇腾论坛_华为云论坛 配置环境:CPU环境 首先下载mindspore,参考官网[MindSpore官网] 安装MindArmour 确认系统环境信息 硬件平台为Ascend、GPU或CPU。 参考MindSpor...

文章首发于笔者华为云,欢迎关注MindArmour 使用(万字详解,gitee同步发表)_MindSpore_昇腾论坛_华为云论坛

配置环境:CPU环境

首先下载mindspore,参考官网[MindSpore官网]

安装MindArmour

确认系统环境信息

  • 硬件平台为Ascend、GPU或CPU。

  • 参考MindSpore安装指南,完成MindSpore的安装。 MindArmour与MindSpore的版本需保持一致。

  • 其余依赖请参见setup.py

安装方式

可以采用pip安装或者源码编译安装两种方式。

pip安装

pip install https://ms-release.obs.cn-north-4.myhuaweicloud.com/{version}/MindArmour/any/mindarmour-{version}-py3-none-any.whl --trusted-host ms-release.obs.cn-north-4.myhuaweicloud.com -i https://pypi.tuna.tsinghua.edu.cn/simple
  • 在联网状态下,安装whl包时会自动下载MindArmour安装包的依赖项(依赖项详情参见setup.py),其余情况需自行安装。

  • {version}表示MindArmour版本号,例如下载1.3.0版本MindArmour时,{version}应写为1.3.0。

源码安装

  1. Gitee下载源码。

    git clone https://gitee.com/mindspore/mindarmour.git
  2. 在源码根目录下,执行如下命令编译并安装MindArmour。

    cd mindarmour
    python setup.py install

验证是否成功安装

执行如下命令,如果没有报错No module named 'mindarmour',则说明安装成功。

python -c 'import mindarmour'

具体操作如下:

image-20211203121312147

如图,最开始没有安装,显示没有mindarmour库

image-20211203121359519

pip命令直接安装。

image-20211203121423684

输入enter之后,没有错误报告,安装正确。

image-20211203121500646

进入python环境,安装正确。

那我们跑一下测试玩玩。

使用NAD算法提升模型安全性

img

img

img

img

开始

刚一开始就报错啦。没事,我们看看信息。

貌似这,暂时CPU还跑不了。

“got device target GPU”。但是仔细分析,我们发现前面这句“support type cpu”。

我们再结合报错信息,只用修改代码中的target即可。

image-20211203122147726

MindSpore的兼容性还是很强的,

稍微调试就好。

果不其然,搞成了target="CPU"就可以了

image-20211203203606566

image-20211203204005181

这就真不错。

image-20211203203710008

经过三轮训练,精确度已经达到97%了

image-20211203203753095

image-20211203204051309

GPU上演示

还没玩够,那我们在gpu上再玩一遍

image-20211203204743935

(差点都忘了自己创建的环境叫什么了,原来叫mindspore1.5-gpu)

遇见的一些问题

GPU运行armour

image-20211203205644470

image-20211203205712776

运行的时候,莫名奇妙出了些小故障,难道python命令出问题了?

image-20211203210029717

原来是c盘满了,我把cuda卸了。看来寒假得重新加一块存储卡...那寒假再跟大家写gpu版本的吧。

完整演示

pycharm加装jupyter

1、安装Jupyter pip install jupyter

2、安装pycharm专业版,然后开始

建立被攻击模型

以MNIST为示范数据集,自定义的简单模型作为被攻击模型。

引入相关包

import os
import numpy as np
from scipy.special import softmax
​
from mindspore import dataset as ds
from mindspore import dtype as mstype
import mindspore.dataset.vision.c_transforms as CV
import mindspore.dataset.transforms.c_transforms as C
from mindspore.dataset.vision import Inter
import mindspore.nn as nn
from mindspore.nn import SoftmaxCrossEntropyWithLogits
from mindspore.common.initializer import TruncatedNormal
from mindspore import Model, Tensor, context
from mindspore.train.callback import LossMonitor
​
from mindarmour.adv_robustness.attacks import FastGradientSignMethod
from mindarmour.utils import LogUtil
from mindarmour.adv_robustness.evaluations import AttackEvaluate
​
context.set_context(mode=context.GRAPH_MODE, device_target="Ascend")
​
LOGGER = LogUtil.get_instance()
LOGGER.set_level("INFO")
TAG = 'demo'

image-20211203220005963

image-20211203220025374

下载文件的时候,会报不信任http,没关系,不用管。

image-20211203220244283

注意,在CPU上运行,设置为target="CPU"

加载数据集

利用MindSpore的dataset提供的MnistDataset接口加载MNIST数据集。

# generate dataset for train of test
def generate_mnist_dataset(data_path, batch_size=32, repeat_size=1,
                           num_parallel_workers=1, sparse=True):
    """
    create dataset for training or testing
    """
    # define dataset
    ds1 = ds.MnistDataset(data_path)
​
    # define operation parameters
    resize_height, resize_width = 32, 32
    rescale = 1.0 / 255.0
    shift = 0.0
​
    # define map operations
    resize_op = CV.Resize((resize_height, resize_width),
                          interpolation=Inter.LINEAR)
    rescale_op = CV.Rescale(rescale, shift)
    hwc2chw_op = CV.HWC2CHW()
    type_cast_op = C.TypeCast(mstype.int32)
​
    # apply map operations on images
    if not sparse:
        one_hot_enco = C.OneHot(10)
        ds1 = ds1.map(operations=one_hot_enco, input_columns="label",
                      num_parallel_workers=num_parallel_workers)
        type_cast_op = C.TypeCast(mstype.float32)
    ds1 = ds1.map(operations=type_cast_op, input_columns="label",
                  num_parallel_workers=num_parallel_workers)
    ds1 = ds1.map(operations=resize_op, input_columns="image",
                  num_parallel_workers=num_parallel_workers)
    ds1 = ds1.map(operations=rescale_op, input_columns="image",
                  num_parallel_workers=num_parallel_workers)
    ds1 = ds1.map(operations=hwc2chw_op, input_columns="image",
                  num_parallel_workers=num_parallel_workers)
​
    # apply DatasetOps
    buffer_size = 10000
    ds1 = ds1.shuffle(buffer_size=buffer_size)
    ds1 = ds1.batch(batch_size, drop_remainder=True)
    ds1 = ds1.repeat(repeat_size)
​
    return ds1
    

image-20211203220532129

建立模型

这里以LeNet模型为例,您也可以建立训练自己的模型。

  1. 定义LeNet模型网络。

    def conv(in_channels, out_channels, kernel_size, stride=1, padding=0):
        weight = weight_variable()
        return nn.Conv2d(in_channels, out_channels,
                         kernel_size=kernel_size, stride=stride, padding=padding,
                         weight_init=weight, has_bias=False, pad_mode="valid")
    ​
    ​
    def fc_with_initialize(input_channels, out_channels):
        weight = weight_variable()
        bias = weight_variable()
        return nn.Dense(input_channels, out_channels, weight, bias)
    ​
    ​
    def weight_variable():
        return TruncatedNormal(0.02)
    ​
    ​
    class LeNet5(nn.Cell):
        """
        Lenet network
        """
        def __init__(self):
            super(LeNet5, self).__init__()
            self.conv1 = conv(1, 6, 5)
            self.conv2 = conv(6, 16, 5)
            self.fc1 = fc_with_initialize(16*5*5, 120)
            self.fc2 = fc_with_initialize(120, 84)
            self.fc3 = fc_with_initialize(84, 10)
            self.relu = nn.ReLU()
            self.max_pool2d = nn.MaxPool2d(kernel_size=2, stride=2)
            self.flatten = nn.Flatten()
    ​
        def construct(self, x):
            x = self.conv1(x)
            x = self.relu(x)
            x = self.max_pool2d(x)
            x = self.conv2(x)
            x = self.relu(x)
            x = self.max_pool2d(x)
            x = self.flatten(x)
            x = self.fc1(x)
            x = self.relu(x)
            x = self.fc2(x)
            x = self.relu(x)
            x = self.fc3(x)
            return x
  2. 训练LeNet模型。利用上面定义的数据加载函数generate_mnist_dataset载入数据。

    mnist_path = "../common/dataset/MNIST/"
    batch_size = 32
    # train original model
    ds_train = generate_mnist_dataset(os.path.join(mnist_path, "train"),
                                      batch_size=batch_size, repeat_size=1,
                                      sparse=False)
    net = LeNet5()
    loss = SoftmaxCrossEntropyWithLogits(sparse=False)
    opt = nn.Momentum(net.trainable_params(), 0.01, 0.09)
    model = Model(net, loss, opt, metrics=None)
    model.train(10, ds_train, callbacks=[LossMonitor()],
                dataset_sink_mode=False)
                

    image-20211203220926643

以下是训练模型的结果

image-20211203220951765

# 2. get test data
ds_test = generate_mnist_dataset(os.path.join(mnist_path, "test"),
                              batch_size=batch_size, repeat_size=1,
                              sparse=False)
inputs = []
labels = []
for data in ds_test.create_tuple_iterator():
 inputs.append(data[0].asnumpy().astype(np.float32))
 labels.append(data[1].asnumpy())
test_inputs = np.concatenate(inputs)
test_labels = np.concatenate(labels)
  1. 测试模型。

    # prediction accuracy before attack
    net.set_train(False)
    test_logits = net(Tensor(test_inputs)).asnumpy()
    
    tmp = np.argmax(test_logits, axis=1) == np.argmax(test_labels, axis=1)
    accuracy = np.mean(tmp)
    LOGGER.info(TAG, 'prediction accuracy before attacking is : %s', accuracy)

    image-20211203221056403

    测试结果中分类精度达到了97%。

对抗性攻击

调用MindArmour提供的FGSM接口(FastGradientSignMethod)。

image-20211203221315683

image-20211203221254692

# attacking
# get adv data
attack = FastGradientSignMethod(net, eps=0.3, loss_fn=loss)
adv_data = attack.batch_generate(test_inputs, test_labels)

# get accuracy of adv data on original model
adv_logits = net(Tensor(adv_data)).asnumpy()
adv_proba = softmax(adv_logits, axis=1)
tmp = np.argmax(adv_proba, axis=1) == np.argmax(test_labels, axis=1)
accuracy_adv = np.mean(tmp)
LOGGER.info(TAG, 'prediction accuracy after attacking is : %s', accuracy_adv)

attack_evaluate = AttackEvaluate(test_inputs.transpose(0, 2, 3, 1),
                                 test_labels,
                                 adv_data.transpose(0, 2, 3, 1),
                                 adv_proba)
LOGGER.info(TAG, 'mis-classification rate of adversaries is : %s',
            attack_evaluate.mis_classification_rate())
LOGGER.info(TAG, 'The average confidence of adversarial class is : %s',
            attack_evaluate.avg_conf_adv_class())
LOGGER.info(TAG, 'The average confidence of true class is : %s',
            attack_evaluate.avg_conf_true_class())
LOGGER.info(TAG, 'The average distance (l0, l2, linf) between original '
            'samples and adversarial samples are: %s',
            attack_evaluate.avg_lp_distance())
LOGGER.info(TAG, 'The average structural similarity between original '
            'samples and adversarial samples are: %s',
            attack_evaluate.avg_ssim())

攻击结果如下:

image-20211203221340678

image-20211203221401541

prediction accuracy after attacking is : 0.052083
mis-classification rate of adversaries is : 0.947917
The average confidence of adversarial class is : 0.803375
The average confidence of true class is : 0.042139
The average distance (l0, l2, linf) between original samples and adversarial samples are: (1.698870, 0.465888, 0.300000)
The average structural similarity between original samples and adversarial samples are: 0.332538

image-20211203221520004

image-20211203221530894

结果如下。

image-20211203221621794

对模型进行FGSM无目标攻击后,模型精度有11%,误分类率高达89%,成功攻击的对抗样本的预测类别的平均置信度(ACAC)为 0.721933,成功攻击的对抗样本的真实类别的平均置信度(ACTC)为 0.05756182,同时给出了生成的对抗样本与原始样本的零范数距离、二范数距离和无穷范数距离,平均每个对抗样本与原始样本间的结构相似性为0.5708779。

对抗性防御

NaturalAdversarialDefense(NAD)是一种简单有效的对抗样本防御方法,使用对抗训练的方式,在模型训练的过程中构建对抗样本,并将对抗样本与原始样本混合,一起训练模型。随着训练次数的增加,模型在训练的过程中提升对于对抗样本的鲁棒性。NAD算法使用FGSM作为攻击算法,构建对抗样本。

防御实现

调用MindArmour提供的NAD防御接口(NaturalAdversarialDefense)。

from mindarmour.adv_robustness.defenses import NaturalAdversarialDefense


# defense
net.set_train()
nad = NaturalAdversarialDefense(net, loss_fn=loss, optimizer=opt,
                                bounds=(0.0, 1.0), eps=0.3)
nad.batch_defense(test_inputs, test_labels, batch_size=32, epochs=10)

# get accuracy of test data on defensed model
net.set_train(False)
test_logits = net(Tensor(test_inputs)).asnumpy()

tmp = np.argmax(test_logits, axis=1) == np.argmax(test_labels, axis=1)
accuracy = np.mean(tmp)
LOGGER.info(TAG, 'accuracy of TEST data on defensed model is : %s', accuracy)

# get accuracy of adv data on defensed model
adv_logits = net(Tensor(adv_data)).asnumpy()
adv_proba = softmax(adv_logits, axis=1)
tmp = np.argmax(adv_proba, axis=1) == np.argmax(test_labels, axis=1)
accuracy_adv = np.mean(tmp)

attack_evaluate = AttackEvaluate(test_inputs.transpose(0, 2, 3, 1),
                                 test_labels,
                                 adv_data.transpose(0, 2, 3, 1),
                                 adv_proba)

LOGGER.info(TAG, 'accuracy of adv data on defensed model is : %s',
            np.mean(accuracy_adv))
LOGGER.info(TAG, 'defense mis-classification rate of adversaries is : %s',
            attack_evaluate.mis_classification_rate())
LOGGER.info(TAG, 'The average confidence of adversarial class is : %s',
            attack_evaluate.avg_conf_adv_class())
LOGGER.info(TAG, 'The average confidence of true class is : %s',
            attack_evaluate.avg_conf_true_class())

image-20211203221958152

在CPU上跑起来了,我已经听到了风扇的声音!

每次跑深度学习模型,都能够听见散热扇呼啸~

数秒后,风扇声音降低,准备查看结果。

防御效果

image-20211203222108855

accuracy of TEST data on defensed model is : 0.981270
accuracy of adv data on defensed model is : 0.813602
defense mis-classification rate of adversaries is : 0.186398
The average confidence of adversarial class is : 0.653031
The average confidence of true class is : 0.184980

使用NAD进行对抗样本防御后,模型对于对抗样本的误分类率降至18%,模型有效地防御了对抗样本。同时,模型对于原来测试数据集的分类精度达98%。

与官网数据对比:

accuracy of TEST data on defensed model is : 0.974259
accuracy of adv data on defensed model is : 0.856370
defense mis-classification rate of adversaries is : 0.143629
The average confidence of adversarial class is : 0.616670
The average confidence of true class is : 0.177374

使用NAD进行对抗样本防御后,模型对于对抗样本的误分类率从95%降至14%,模型有效地防御了对抗样本。同时,模型对于原来测试数据集的分类精度达97%。

image-20211203233657902

开源代码

亲爱的朋友,我已将本文中MindArmour的实操代码开源到gitee,代码已经在CPU上调试通过,欢迎大家下载使用,亲手调试后会有更加深入的理解。

链接:MindSporeArmour: Show details of how to use MindSpore Armour.

文章来源: blog.csdn.net,作者:irrationality,版权归原作者所有,如需转载,请联系作者。

原文链接:blog.csdn.net/weixin_54227557/article/details/121757633

【版权声明】本文为华为云社区用户转载文章,如果您发现本社区中有涉嫌抄袭的内容,欢迎发送邮件进行举报,并提供相关证据,一经查实,本社区将立刻删除涉嫌侵权内容,举报邮箱: cloudbbs@huaweicloud.com
  • 点赞
  • 收藏
  • 关注作者

评论(0

0/1000
抱歉,系统识别当前为高风险访问,暂不支持该操作

全部回复

上滑加载中

设置昵称

在此一键设置昵称,即可参与社区互动!

*长度不超过10个汉字或20个英文字符,设置后3个月内不可修改。

*长度不超过10个汉字或20个英文字符,设置后3个月内不可修改。