基于RF算法的侧信道攻击方法研究

doi:10.3969/j.issn.1671-1122.2022.01.003

摘要/Abstract

摘要：

目前,随机森林（RF）算法在侧信道分析领域的潜力还没有得到充分利用。文章提出一种基于RF算法的侧信道攻击方法,分别从输入数据处理和参数控制两方面进行模型优化,在特征点选择和RF算法参数调优两方面进行改进。对于高维数据,首先使用SOST相关系数法选出100个特征点,然后对RF算法的各参数进行调优。实验结果表明,与采用默认参数值的基于RF算法的侧信道攻击相比,该方法的攻击成功率显著提高,模型的泛化能力也有一定程度的提高。

关键词: 机器学习, 侧信道攻击, RF算法, 模型选择

Abstract:

At present, the full potential of random forest(RF) algorithm in the field of side channel analysis has not been fully utilized. This paper proposed a side channel attack based on RF algorithm, which optimized the model from input data processing and parameter control, and improved it from feature point selection and RF algorithm parameter optimization. For high-dimensional data, the SOST correlation coefficient method was used to select 100 feature points, and then optimize the parameters of RF algorithm. The results show that compared with the RF algorithm directly based on the default parameter value, the attack success rate of this method is dramatically, and the generalization ability of the model is also improved by a certain extent.

Key words: machine learning, side channel attack, RF algorithm, model selection

中图分类号:

TP309

段晓毅, 李邮, 令狐韫行, 胡荣磊. 基于RF算法的侧信道攻击方法研究[J]. 信息网络安全, 2022, 22(1): 19-26.

DUAN Xiaoyi, LI You, LINGHU Yunxing, HU Ronglei. Research on the Method of Side Channel Attack Based on RF Algorithm[J]. Netinfo Security, 2022, 22(1): 19-26.

图/表 24

表1

图1

表2

图2

图3

表3

图4

表4

图5

表5

图6

表6

图7

表7

图8

表8

图9

表9

表10

图10

图11

表11

表12

表13

参考文献 13

[1]	HOSPODAR G, GIERLICHS B, MULDER E D, et al. Machine Learning in Side-channel Analysis: A First Study[J]. Journal of Cryptographic Engineering, 2011, 1(4):293-302. doi: 10.1007/s13389-011-0023-x URL
[2]	LERMAN L, BONTEMPI G, MARKOWITCH O. A Machine Learning Approach Against a Masked AES[J]. Journal of Cryptographic Engineering, 2014, 5(2):123-139. doi: 10.1007/s13389-014-0089-3 URL
[3]	HOSPODAR G, GIERLICHS B, MULDER E D, et al. Machine Learning in 420 Side-channel Analysis: a First Study[J]. Journal of Cryptographic Engineering, 2011, 1(4):293-302. doi: 10.1007/s13389-011-0023-x URL
[4]	HEUSER A, ZOHNER M. Constructive Side-channel Analysis and Secure Design[M]. Heidelberg: Intelligent Homicide, 2012.
[5]	WANG Feiyu, CHEN Botao, LIU Jianfeng, et al. Combined Attack of Selective Plaintext and Second-order DPA against Masked AES[J]. Journal of Cryptography, 2014, 1(6):525-536.
[6]	TIMON B. Non-profiled Deep Learning-based Side-channel Attacks with Sensitivity Analysis[EB/OL]. https://doi.org/10.13154/tches.v2019.i2.107-131, 2021-03-11.
[7]	KIM J, PICEK S, HEUSER A, et al. Make Some Noise. Unleashing the Power of Convolutional Neural Networks for Profiled Side-channel Analysis[EB/OL]. https://doi.org/10.13154/tches.v2019.i3.148-179, 2021-03-12.
[8]	BENADJILA R, PROUFF E, STRULLU R, et al. Deep Learning for Side-channel Analysis and Introduction to ASCAD Database[J]. Journal of Cryptographic Engineering, 2020, 10(2):163-188. doi: 10.1007/s13389-019-00220-8 URL
[9]	MA Xiangliang, WANG Hong, LI Bing, et al. Chip Backdoor Instruction Analysis Method Based on Energy Analysis Technology[J]. Chinese Journal of Electronics, 2019, 47(3):686-691.
	马向亮, 王宏, 李冰, 等. 基于能量分析技术的芯片后门指令分析方法[J]. 电子学报, 2019, 47(3):686-691. doi: 10.3969/j.issn.0372-2112.2019.03.024
[10]	CAI Xiaomin, LI Renfa, KUANG Shijie, et al. An Energy Trace Compression Method for Differential Power Analysis Attack[EB/OL]. https://ieeexplore.ieee.org/document/9091021, 2021-03-12.
[11]	DUAN Xiaoyi, CHEN Dong, FAN Xiaohong, et al. Research and Implementation on Power Analysis Attacks for Unbalanced Data[EB/OL]. https://doi.org/10.1155/2020/5695943, 2020-05-22.
[12]	SHI Rongxi, YANG Borui, JIANG Yang, et al. Energy Attack: on Transferring Adversarial Examples[EB/OL]. https://arxiv.org/abs/2109.04300v1, 2021-03-14.
[13]	CHARI S, RAO J R. Pankaj Rohatgi: Template Attacks[EB/OL]. https://www.researchgate.net/publication/244437815_Pankaj_Rohatgi_Template_attacks, 2021-03-15.

参数	含义	默认值
n_estimators	随机森林中树的棵数	10
max_features	判断条件可选的特征数量	None
max_depth	树的最大深度	None
min_sample_split	最小划分样本数	2
max_leaf_nodes	最大叶子节点数	None
min_impurity_split	节点的最小不纯度	0
random_state	随机状态	None

参数	参数对模型评估性能的影响	影响程度
n_estimators	n_estimators增加不影响单个模型的复杂度	非常大
max_depth	max_depth增加模型更简单,且向图像左边移动	大
min_samples_leaf	min_samples_leaf增加模型更简单,且向图像左边移动	中等
min_samples_split	min_samples_leaf增加模型更简单,且向图像左边移动	中等
max_features	既能使模型更简单,也能使模型更复杂	弱

n_estimators	模型准确率
100	0.938444
200	0.941667
300	0.943778
400	0.944333
500	0.944444
600	0.945222
700	0.945778
800	0.945111
900	0.944778
1000	0.944444

n_estimators	模型准确率
600	0.945222
650	0.944778
700	0.945778
750	0.945000
800	0.945111

n_estimators	模型准确率
690	0.945333
700	0.945778
710	0.945778
720	0.945556