Research Article
Power Analysis Attack Based on GA-Based Ensemble Learning
@INPROCEEDINGS{10.1007/978-3-031-56580-9_19, author={Xiaoyi Duan and Ye Huang and Yuting Wang and Yu Gu and Jianmin Tong and Zunyang Wang and Ronglei Hu}, title={Power Analysis Attack Based on GA-Based Ensemble Learning}, proceedings={Digital Forensics and Cyber Crime. 14th EAI International Conference, ICDF2C 2023, New York City, NY, USA, November 30, 2023, Proceedings, Part I}, proceedings_a={ICDF2C}, year={2024}, month={4}, keywords={Index Terms---Power Analysis Attack Machine Learning Hyperparameter Search Ensemble Learning Genetic Algorithm}, doi={10.1007/978-3-031-56580-9_19} }- Xiaoyi Duan
Ye Huang
Yuting Wang
Yu Gu
Jianmin Tong
Zunyang Wang
Ronglei Hu
Year: 2024
Power Analysis Attack Based on GA-Based Ensemble Learning
ICDF2C
Springer
DOI: 10.1007/978-3-031-56580-9_19
Abstract
Perin et al. proposed the random ensemble leaning method. This method generates multiple neural network models, randomly sets the parameters of the models, then the models are integrated to perform power analysis attacks. Compared with single model, this method is more efficient, which performs very well in enhancing the performance of side-channel attacks. However, Perin’s solution does not solve the problem of combinatorial optimization during ensemble and relatively requires more neural networks to be integrated. This paper proposes a GA-based Ensemble Learning method, which generates multiple neural network models, then obtains the optimal parameters of the models through the genetic algorithm to solve the optimal combination problem of the integrated neural network, and finally use the network with optimal parameters for power analysis attacks. Compared with Perin’s method, the proposed method needs less neural network ensemble to achieve better results. Compared with Perin’s random ensemble learning method on three data sets (ASCADf, ASCADr, CHES CTF) respectively, in order to achieve the same attack effect, GA-based Ensemble method reduces 10 models in ensemble scale than Perin’s Random Ensemble learning method. The GA-based ensemble learning method can further improve the attack performance of the ensemble and reduce the scale of the ensemble, thereby saving the training cost.
