Secure principal component analysis in multiple distributed nodes

Abstract

Privacy preservation becomes an important issue in recent big data analysis, and many secure multiparty computations have been proposed for the purpose of privacy preservation in the environment of distributed nodes. As a secure multiparty computations of principal component analysis (PCA), in this paper, we propose S‐PCA, which compute PCA securely among the distributed nodes. PCA is widely used in many applications including time‐series analysis, text mining, and image compression. In general, we compute PCA after concentrating all data in a single server, but this approach discloses data privacy of each node. In contrast, the proposed S‐PCA computes PCA without disclosing the sensitive data of individual nodes. In S‐PCA, the nodes share non‐sensitive mean vectors first and compute covariance matrices and PCA securely using the shared mean vectors. In this paper, we formally prove the correctness and secureness of S‐PCA and apply it to an application of secure similar document detection. Experimental results show that the performance of S‐PCA is slightly worse than that of PCA due to guarantee of secureness, but it significantly improves the performance of secure similar document detection by up to two orders of magnitudes. Copyright © 2016 John Wiley & Sons, Ltd.

Federated horizontally partitioned principal component analysis for biomedical applications

“…Several articles discuss privacy-preserving PCA or power iteration in a federated setting via encryption and secure multiparty computation techniques ( Al-Rubaie et al , 2017 ; Cho et al , 2018 ; Pathak and Raj, 2011 ; Rathee et al , 2018 ; Won et al , 2016 ) or differential privacy (DP; Balcan et al , 2016 ; Hardt and Price, 2014 ; Imtiaz and Sarwate, 2018 ; Wang and Morris Chang, 2019 ). A few articles ( Liu et al , 2020 ; Won et al , 2016 ) assume that the aggregated covariance matrix is private. The authors of Pathak and Raj (2011) assume that the aggregated eigenvector updates are private.…”

“…A means to significantly reduce the transmission costs is to approximate the local subspaces and send these to the aggregator. In this case, a local singular value decomposition (SVD) is computed and the top-k eigenspace is sent to the aggregator, where k is fixed but arbitrary ( Al-Rubaie et al , 2017 ; Fan et al , 2019 ; Imtiaz and Sarwate, 2018 ; Liu et al , 2018 ; Qu et al , 2002 ; Wang and Morris Chang, 2019 ; Won et al , 2016 ). More precisely, in these algorithms, the local subspace is computed at each site and sent to the aggregator ( Algorithm 1 , Lines 2 and 3).…”

PD-ML-Lite: Private Distributed Machine Learning from Lightweight Cryptography

“…However a consistent issue is that in addition to V k , Σ k is typically revealed. In [63,14] the methods essentially reduce to sharing X T X = V Σ 2 V T securely, and as a result each party learns Σ. In [30] a method is proposed based on the QR decomposition that allows parties to only share Σ if they choose to, however it is only developed for M = 2 parties.…”

“…We also show that PD-SVD is more private than Σ-revealing SVD algorithms that use secure multiparty computation but expose Σ in addition to V k . [63,14].…”

All-in-One Framework for Detection, Unpacking, and Verification for Malware Analysis

“…According to the 2017 AV-Test security report [1], about six billion malwares are used annually in DDoS (distributed denial of service), spam mails, and APT (advanced persistent threat). In addition, due to the advent of new malwares exploiting analysis avoidance techniques, there have been many research e orts on personal information protection, malicious code detection, and malware analysis technology [2][3][4][5][6][7][8][9][10]. Among the analysis avoidance techniques, packing is the most common one used to hide malware.…”