[논문]Improve the Performance of Semi-Supervised Side-channel Analysis Using HWFilter Method

Hong Zhang; Lang Li; Di Li

doi:10.3837/tiis.2024.03.012

[국내논문] Improve the Performance of Semi-Supervised Side-channel Analysis Using HWFilter Method

KSII Transactions on internet and information systems : TIIS, v.18 no.3, 2024년, pp.738 - 754

Hong Zhang (College of Computer Science and Technology, Hengyang Normal University) , Lang Li (College of Computer Science and Technology, Hengyang Normal University) , Di Li (College of Computer Science and Technology, Hengyang Normal University)

Abstract ▼ AI-Helper

Side-channel analysis (SCA) is a cryptanalytic technique that exploits physical leakages, such as power consumption or electromagnetic emanations, from cryptographic devices to extract secret keys used in cryptographic algorithms. Recent studies have shown that training SCA models with semi-supervised learning can effectively overcome the problem of few labeled power traces. However, the process of training SCA models using semi-supervised learning generates many pseudo-labels. The performance of the SCA model can be reduced by some of these pseudo-labels. To solve this issue, we propose the HWFilter method to improve semi-supervised SCA. This method uses a Hamming Weight Pseudo-label Filter (HWPF) to filter the pseudo-labels generated by the semi-supervised SCA model, which enhances the model's performance. Furthermore, we introduce a normal distribution method for constructing the HWPF. In the normal distribution method, the Hamming weights (HWs) of power traces can be obtained from the normal distribution of power points. These HWs are filtered and combined into a HWPF. The HWFilter was tested using the ASCADv1 database and the AES_HD dataset. The experimental results demonstrate that the HWFilter method can significantly enhance the performance of semi-supervised SCA models. In the ASCADv1 database, the model with HWFilter requires only 33 power traces to recover the key. In the AES_HD dataset, the model with HWFilter outperforms the current best semi-supervised SCA model by 12%.

주제어

표/그림 (16)

그림 Fig. 1. The process of semi-supervised learning
그림 Fig. 2. The process of using the self-training method in SCA.
그림 Fig. 3. The HWFilter method.
그림 Fig. 4. One hundred thousand SNR values are related to the intermediate values, the red dot represents the index of power point with the maximum SNR value.
그림 Fig. 5. The probability distribution graph of 100,000 power points
그림 Fig. 6. The process of using HWPF.
그림 Fig. 7. The structure of the semi-supervised SCA model.
표 Table 1. The hyperparametric search spaces of SCA models.
그림 Fig. 8. About the PCC of 1400 power points in ASCAD_f.
그림 Fig. 9. About the PCC of 1400 power points in ASCAD_r.
그림 Fig. 10. The power values of 700 power points in the AES_HD dataset.
그림 Fig. 11. The GE results of different ID models on ASCAD_f.
그림 Fig. 12. The GE results of different ID models on ASCAD_r.
표 Table 2. GE results on AES_HD dataset (-if not reached).
표 Table 3. GE results on ASCAD_f (-if not reached).
표 Table 4. GE results on ASCAD_r (-if not reached).

AI 본문요약
AI-Helper

제안 방법

The ID model is the most widely used semi-supervised SCA model. In this experiment, we tested the performance of HWFilter using the ID model. We first compared the performance of ID model with HWPF and the performance of ID model without HWPF by using the ASCAD_f.
(1) We propose a HWFilter method, which is designed to enhance the performance of semi-supervised SCA models. This method uses a Hamming Weight Pseudo-label Filter(HWPF) to filter out the pseudo-labels generated by the semi-supervised SCA model, thereby reducing the incorrect pseudo-labels involved in model training.

성능/효과

(1) We propose a HWFilter method, which is designed to enhance the performance of semi-supervised SCA models. This method uses a Hamming Weight Pseudo-label Filter(HWPF) to filter out the pseudo-labels generated by the semi-supervised SCA model, thereby reducing the incorrect pseudo-labels involved in model training.
(2) We propose a normal distribution method for constructing HWPFs. The method can compute the HWs for each power consumption trace from the normal distribution of power consumption points.
(3) We used the ASCADv1 database and the AES_HD dataset to test the HWFilter method. The experimental results show that the performance of the semi-supervised SCA model trained using the HWFilter method is higher than the performance of other SCA models.
In the profiling phase of Fig. 2, pseudo-labels contain correct labels and incorrect labels. Correct labels can speed up the convergence of the model and improve the performance of the model.
In Fig. 3, the HWFilter method consists of a construct HWPF phase and a Use HWPF phase. In the construct HWPF phase, we use the normal distribution method to construct a HWPF.
In Fig. 6, the pre-trained model generates some pseudo-labels. These pseudo-labels are first filtered through a confidence interval and then filtered using HWPF.
We then tested the performance of ID model with HWPF on the ASCAD_r. Although the ASCAD_r is more difficult than the ASCAD_f dataset, ID model with HWPF still performs well. The ID model with HWPF can use fewer power traces to recover key when the number of power traces with intermediate values is reduced.
Table 2 shows the performance of HWPF on the AES_HD dataset. Although the maximum PCC of the AES_HD dataset is only 0.041, HWPF can still improve the performance of the ID model. When the number of labeled power traces is only 10,000, the performance of the model with HWFilter is 12% higher than that of Picek et al.
The performance of the semi-supervised SCA model with HWPF is higher than that of the supervised SCA model for a smaller number of labeled power traces. As the number of labeled power consumption trajectories increases, the performance of the semi-supervised SCA model with HWPF and the supervised SCA model converge to be equal.
If there are not enough labeled power traces, it is difficult for the attacker to obtain a SCA model with better performance. Based on this hypothesis, The main challenge of SCA with semi-supervised learning is how to use very few labeled power traces to improve the possibility of recovering keys. Some researchers have employed techniques such as self-training and graph-based learning to address this challenge [15-17].
Although this type of DL-SCA exhibits significant advantages over traditional non-profiling SCAs such as DPA [8] and CPA [9], it still presents certain gaps when compared to DL-SCA with supervised learning. DL-SCA with semi-supervised learning overcomes the limitations of the previous two types of DL-SCA, enabling key recovery even when labeled power traces are scarce.
In this paper, a HWFilter method is proposed to improve the performance of semi-supervised SCA models. This method can reduce the number of erroneous pseudo-labels and accelerate the convergence of the model.
0001 and the NLLLoss function, respectively. The batch size and training epoch are 200 and 300, respectively. The structure of the model is shown in Fig.
(3) We used the ASCADv1 database and the AES_HD dataset to test the HWFilter method. The experimental results show that the performance of the semi-supervised SCA model trained using the HWFilter method is higher than the performance of other SCA models.
11 shows the GE of different ID models on ASCAD_f. The performance of ID models with HWPF is better than those without HWPF when the number of power traces with intermediate values is very small.
Table 3 shows the performance of HWPF on the ASCAD_f. The performance of the semi-supervised SCA model with HWPF is higher than that of the supervised SCA model for a smaller number of labeled power traces. As the number of labeled power consumption trajectories increases, the performance of the semi-supervised SCA model with HWPF and the supervised SCA model converge to be equal.
Finally, we conducted semi-supervised SCA experiments using the ASCADv1 database and the AES_HD dataset. These experimental results show that HWFilter can promote the training of semi-supervised SCA models. In particular, when the number of labeled power traces is small, the model trained using the HWFilter method outperforms the model trained using the supervised method.
This algorithm has been able to improve accuracy by about 20% [18]. These studies show that SCA with semi-supervised learning can successfully recover keys even with a limited number of labeled power traces.

후속연구

This allows the attacker to obtain only a small number of labeled power traces during the profiling phase. If there are not enough labeled power traces, it is difficult for the attacker to obtain a SCA model with better performance. Based on this hypothesis, The main challenge of SCA with semi-supervised learning is how to use very few labeled power traces to improve the possibility of recovering keys.
In this paper, we present some methods to solve the problem of erroneous pseudo-labels that hinder the training of semi-supervised SCA models. More research is needed in the future to investigate how semi-supervised learning can further promote the development of the SCA field.

참고문헌 (26)

Picek S, Perin G, Mariot L, et al., "Sok: Deep learning-based physical side-channel analysis," ACM Computing Surveys, vol. 55, no. 11, pp. 1-35, 2023.

상세보기
Rijsdijk J, Wu L, Perin G, et al., "Reinforcement learning for hyperparameter tuning in deep learning-based side-channel analysis," IACR Transactions on Cryptographic Hardware and Embedded Systems, vol. 2021, no. 3, pp. 677-707, 2021.
Ito A, Ueno R, Homma N, "Perceived information revisited: New metrics to evaluate success rate of side-channel attacks," IACR Transactions on Cryptographic Hardware and Embedded Systems, vol. 2022, no. 4, pp. 228-254, 2022.
Yap T, Benamira A, Bhasin S, et al., "Peek into the Black-Box: Interpretable Neural Network using SAT Equations in Side-Channel Analysis," IACR Transactions on Cryptographic Hardware and Embedded Systems, vol. 2023, no. 2, pp. 24-53, 2023.
Ramezanpour K, Ampadu P, Diehl W, "SCAUL: Power side-channel analysis with unsupervised learning," IEEE Transactions on Computers, vol. 69, no. 1, pp. 1626-1638, 2020.
Cao, P., Zhang, C., Lu, X., "Cross-Device Profiled Side-Channel Attack with Unsupervised Domain Adaptation," IACR Transactions on Cryptographic Hardware and Embedded Systems, vol. 2021, no. 4, pp. 27-56, 2021.
Kulow A, Schamberger T, Tebelmann L, et al., "Finding the needle in the haystack: Metrics for best trace selection in unsupervised side-channel attacks on blinded RSA," IEEE Transactions on Information Forensics and Security, vol. 16, no. 3, pp. 3254-3268, 2021.
Cheng J, Liu W, Sun N, et al., "A machine learning low-dropout regulator-assisted differential power analysis attack countermeasure with voltage scaling," International Journal of Circuit Theory and Applications, vol. 51, no. 7, pp. 3105-3117, 2023.

상세보기
Do N T, Hoang V P, Pham C K, "Low Complexity Correlation Power Analysis by Combining Power Trace Biasing and Correlation Distribution Techniques," IEEE Access, vol. 10, pp. 17578- 7589, 2022.

상세보기
Cai J, Hao J, Yang H, et al., "A review on semi-supervised clustering," Information Sciences, vol. 632, pp. 164-200, 2023.

상세보기
Van Engelen J E, Hoos H H, "A survey on semi-supervised learning," Machine learning, vol. 109, no. 2, pp. 373-440, 2020.
Taha K, "Semi-supervised and un-supervised clustering: A review and experimental evaluation," Information Systems, vol. 114, pp. 102178, 2023.

상세보기
Chen Y, Tan X, Zhao B, et al., "Boosting Semi-Supervised Learning by Exploiting All Unlabeled Data," in Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, IEEE, Oxford, UK, pp. 7548-7557, 2023.
Picek S, Heuser A, Jovic A, et al., "Improving side-channel analysis through semi-supervised learning," in Proc. of International Conference on Smart Card Research and Advanced Applications, Springer, Montpellier, France, pp. 35-50, 2019.
Batina L, Djukanovic M, Heuser A, et al., "It started with templates: The future of profiling in side-channel analysis," in Security of Ubiquitous Computing Systems: Selected Topics, vol. 3, 2021, pp. 133-145.
Panoff M, Yu H, Shan H, et al., "A Review and Comparison of AI-enhanced Side Channel Analysis," ACM Journal on Emerging Technologies in Computing Systems (JETC), vol. 18, no. 3, pp. 1-20, 2022.
Hettwer B, Gehrer S, Guneysu T, "Applications of machine learning techniques in side-channel attacks: a survey," Journal of Cryptographic Engineering, vol. 10, pp. 135-162, 2020.

상세보기
Biao Liu, Zhao Ding, Yang Pan, Jiali Li, and Huamin Feng, "Side-channel attacks based on collaborative learning," in Proc. of Data Science: Third International Conference of Pioneering Computer Scientists, Engineers and Educators, Springer, Changsha, China, pp. 549-557, 2017.
Tang H, Jia K, "Towards discovering the effectiveness of moderately confident samples for sem-isupervised learning," in Proc. of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, IEEE, New Orleans, USA, pp. 14658-14667, 2022.
Huan P T, Thong P H, Tuan T M, et al., "TS3FCM: trusted safe semi-supervised fuzzy clustering method for data partition with high confidence," Multimedia Tools and Applications, vol. 81, no. 9, pp. 12567-12598, 2022.

상세보기
Wang W, Zhang M L, "Semi-supervised partial label learning via confidence-rated margin maximization," in Proc. of the 34th International Conference on Neural Information Processing Systems, pp. 6982-6993, 2020.
Stjepan Picek, Annelie Heuser, Alan Jovic, Shivam Bhasin, and Francesco Regazzoni, "The curse of class imbalance and conflicting metrics with machine learning for side-channel evaluations," IACR Transactions on Cryptographic Hardware and Embedded Systems, vol. 2019, no. 1, pp. 209-237, 2018.
Zhang H, Yang W, "Template Attack Assisted Linear Cryptanalysis on Outer Rounds Protected DES Implementations," The Computer Journal, vol. 66, no. 6, pp. 1434-1451, 2023.

상세보기
Ryad Benadjila, "Emmanuel Prouff, Remi Strullu, Eleonora Cagli, and Cecile Dumas, "Deep learning for side-channel analysis and introduction to ascad database," Journal of Cryptographic Engineering, vol. 10, no. 2, pp. 163-188, 2020.
Huanhuan Ran, Shiping Wen, Qian Li, Yuting Cao, Kaibo Shi, and Tingwen Huang, "Compact and stable memristive visual geometry group neural network," IEEE Transactions on Neural Networks and Learning Systems, vol. 34, no 2, pp. 987-998, 2023.

상세보기
Alexander Ly, Maarten Marsman, and Eric-Jan Wagenmakers, "Analytic posteriors for Pearson's correlation coefficient," Statistica Neerlandica, vol. 72, no. 1, pp. 4-13, 2018.

상세보기

표제어: PCR

동의어: Packet Collision Rate

용어 설명 출처 목록 (6)

용어 설명: PCR은 세균 특이성이 있는 primer를 이용하여 적은 수의 세균이 있을지라도 쉽게 검출할 수 있는 유용한 방법이며, 이를 이용하여 구강 내 치면세균막이나 타액에서 직접 세균을 검출할 수 있게 되었다[8].

내보내기 구분	파일저장 인쇄 메일전송
구성항목	기본정보 상세정보 관리번호, 논문명, 저널/프로시딩명, 저자 , 발행년, 권, 호, 시작페이지, 끝페이지, 발행기관 관리번호, 논문명, 대등논문명, 저자 , 저널/프로시딩명, 발행기관, 발행년, 발행언어, 권, 호, 시작페이지, 끝페이지, ISBN, ISSN, 주제분야, 키워드, 초록(한글), 초록(영문), 저자(소속기관)
저장형식	Text(ASCII format) Excel format RefWorks Direct Export RIS format (for Reference Manager, ProCite, EndNote), Scholar's Aids, Mendeley
메일정보	받는사람 (필수) @ 보내는사람 (선택) @ 제목 내용 KISTI 검색결과 이메일 서비스
안내	총 건의 자료가 검색되었습니다. 다운받으실 자료의 인덱스를 입력하세요. (1-10,000) 검색결과의 순서대로 최대 10,000건 까지 다운로드가 가능합니다. 데이타가 많을 경우 속도가 느려질 수 있습니다.(최대 2~3분 소요) 다운로드 파일은 UTF-8 형태로 저장됩니다. 파일의 내용이 제대로 보이지 않을실 때는 웹브라우저 상단의 보기 -> 인코딩 -> 자동선택 여부를 확인하십시오. ~ Text(ASCII format) Excel format

연합인증