当前位置:网站首页>将恶意软件嵌入到神经网络中
将恶意软件嵌入到神经网络中
2022-07-02 06:26:00 【MezereonXP】
将恶意软件嵌入到神经网络中
这次介绍一篇名为“EvilModel: Hiding Malware Inside of Neural Network Models ”的文章。
该文章主要描述了一种隐式的传播恶意软件的方法,通过对神经网络的权重进行修改实现恶意软件的传播。

本质上,是选取模型中的某些“冗余”的层,对其中的神经元的权重进行替换,对于每一个权重,替换其最后的3位比特位作为恶意软件的某个3个比特,进而实现恶意软件的隐式传输。
这里所提及的冗余,其实是通过测试,查看对正确率的影响程度,选择那些对正确率影响程度比较小的层。

如上图所示,横轴是替换神经元的个数,纵轴是修改后模型的正确率。
这里有一些观察:
- 随着修改的神经元个数越多,正确率受到的影响就越大。
- 更加靠近输入层的神经元,更加适合嵌入恶意软件,影响更小
由于对神经网络权重的修改会降低其原本的正确率,所以作者也探究了重训练的效果,如下图所示:

其中,BR表示在重训练之前 (Before),AR则表示在重训练之后 (After)
可以看到,没有带BN (BatchNorm) 层的网络,通过重训练无法回复精度,而带有BN层的则可以通过重训练恢复精度。
边栏推荐
- Generate random 6-bit invitation code in PHP
- The difference and understanding between generative model and discriminant model
- 聊天中文语料库对比(附上各资源链接)
- [binocular vision] binocular stereo matching
- Faster-ILOD、maskrcnn_ Benchmark installation process and problems encountered
- Play online games with mame32k
- One book 1078: sum of fractional sequences
- Feature Engineering: summary of common feature transformation methods
- What if the laptop task manager is gray and unavailable
- Proof and understanding of pointnet principle
猜你喜欢

【Cascade FPD】《Deep Convolutional Network Cascade for Facial Point Detection》

iOD及Detectron2搭建过程问题记录

Traditional target detection notes 1__ Viola Jones

Faster-ILOD、maskrcnn_ Benchmark installation process and problems encountered

【MobileNet V3】《Searching for MobileNetV3》

Timeout docking video generation

【Random Erasing】《Random Erasing Data Augmentation》

MoCO ——Momentum Contrast for Unsupervised Visual Representation Learning

基于onnxruntime的YOLOv5单张图片检测实现

常见的机器学习相关评价指标
随机推荐
超时停靠视频生成
(15) Flick custom source
TimeCLR: A self-supervised contrastive learning framework for univariate time series representation
【Sparse-to-Dense】《Sparse-to-Dense:Depth Prediction from Sparse Depth Samples and a Single Image》
iOD及Detectron2搭建过程问题记录
【AutoAugment】《AutoAugment:Learning Augmentation Policies from Data》
Use matlab to realize: chord cut method, dichotomy, CG method, find zero point and solve equation
Determine whether the version number is continuous in PHP
【MobileNet V3】《Searching for MobileNetV3》
【Hide-and-Seek】《Hide-and-Seek: A Data Augmentation Technique for Weakly-Supervised Localization xxx》
【Hide-and-Seek】《Hide-and-Seek: A Data Augmentation Technique for Weakly-Supervised Localization xxx》
Pointnet understanding (step 4 of pointnet Implementation)
Huawei machine test questions-20190417
Yolov3 trains its own data set (mmdetection)
TimeCLR: A self-supervised contrastive learning framework for univariate time series representation
机器学习理论学习:感知机
open3d学习笔记四【表面重建】
Comparison of chat Chinese corpus (attach links to various resources)
Calculate the total in the tree structure data in PHP
yolov3训练自己的数据集(MMDetection)