面向物联网安全的模型攻击研究  

Research on Model Attacks for IoT Security

作　　者：汪浩博 陈博[2] WANG Hao-bo;CHEN Bo(School of Informatics Science and Technology,Zhejiang SCI-TECH University,Hangzhou 310018,China;College of Electrical and Electronic Engineering,Wenzhou University,Wenzhou 325035,China)

机构地区：[1]浙江理工大学信息学院,浙江杭州310018 [2]温州大学电气与电子工程学院,浙江温州325035

出　　处：《电脑知识与技术》2020年第32期57-58,67,共3页Computer Knowledge and Technology

基　　金：温州市基础性工业科技项目(编号:G20180010),名称:面向集成电路IP核的行为算法级硬件混淆研究。

摘　　要：物理不可克隆函数以其轻量化和低功耗等特点,在系统认证和密钥生成等面向物联网安全领域得到广泛应用。通过对典型的强物理不可克隆函数进行建模,结合人工智能和机器学习的手段,采用经典的逻辑回归分析,对强物理不可克隆函数进行模型攻击。在收集一定数量的样本作为训练集的情况下,对特定的测试集进行验证。结果显示,随着训练数据的增多,预测正确率快速升高,并最终稳定在一个较高的水平。Physical unclonable functions are widely used in IoT security fields,such as system authentication and secret key generation,due to their lightweight and low power consumption.This article uses classic logistics regression analysis to model attacks by modeling the typical strong physical unclonable functions and combining artificial intelligence and machine learning.In the case of collecting a certain number of samples as the training set,verify the specific test set.The results show that with the increase of training data,the prediction accuracy rate increases rapidly and finally stabilizes at a higher level.

关 键 词：物联网 机器学习 逻辑回归 模型攻击 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]