An Experimental Study on Attacking Homogeneous Averaging Processes via Side Channel Attacks
2024-5-27 12:8:33 Author: www.ndss-symposium.org(查看原文) 阅读量:1 收藏

Olsan Ozbay (Dept. ECE, University of Maryland), Yuntao Liu (ISR, University of Maryland), Ankur Srivastava (Dept. ECE, ISR, University of Maryland)

Electromagnetic (EM) side channel attacks (SCA) have been very powerful in extracting secret information from hardware systems. Existing attacks usually extract discrete values from the EM side channel, such as cryptographic key bits and operation types. In this work, we develop an EM SCA to extract continuous values that are being used in an averaging process, a common operation used in federated learning. A convolutional neural network (CNN) framework is constructed to analyze the collected EM data. Our results show that our attack is able to distinguish the distributions of the underlying data with up to 93% accuracy, indicating that applications previously considered as secure, such as federated learning, should be protected from EM side-channel attacks in their implementation.


文章来源: https://www.ndss-symposium.org/ndss-paper/auto-draft-500/
如有侵权请联系:admin#unsafe.sh