One-Pixel Attack for Continuous-Variable Quantum Key Distribution Systems

Abstract

Deep neural networks (DNNs) have been employed in continuous-variable quantum key distribution (CV-QKD) systems as attacking detection portions of defense countermeasures. However, the vulnerability of DNNs leaves security loopholes for hacking attacks, for example, adversarial attacks. In this paper, we propose to implement the one-pixel attack in CV-QKD attack detection networks and accomplish the misclassification on a minimum perturbation. This approach is based on the differential evolution, which makes our attack algorithm fool multiple DNNs with the minimal inner information of target networks. The simulation and experimental results show that, in four different CV-QKD detection networks, 52.8%, 26.4%, 21.2%, and 23.8% of the input data can be perturbed to another class by modifying just one feature, the same as one pixel for an image. We carry out this success rate in the context of the original accuracy reaching up to nearly 99% on average. Further, by enlarging the number of perturbed features, the success rate can be raised to a satisfactory higher level of about 80%. According to our experimental results, most of the CV-QKD detection networks can be deceived by launching one-pixel attacks.

Dictionary Learning Based Scheme for Adversarial Defense in Continuous-Variable Quantum Key Distribution

“…What is worse is that recent studies have shown that adversarial attacks can be applied to the real world by those possessing ulterior motivates, which pose a serious threat to areas requiring high information security, for example, autonomous driving [ 26 ]. Similarly, for the CVQKD system using neural networks for attack detection and classification, adversarial attacks can also affect the CVQKD system, which is proved by our simulation experiments [ 27 ].…”

Adversarial Perturbation Elimination with GAN Based Defense in Continuous-Variable Quantum Key Distribution Systems

“…Input: input data X, perturbation ε, number of data N, attack type D, parameter θ, λ Initialization: θ, λ and ε For i ← 0 to N do (x i , y i ) ∼ X // Sampling from normal distribution x * i ← G(O(x i ), y LL , ε) // Construct adversarial samples through ( 14) and ( 15) Get the adversarial loss function through (13) θ ← θ − η • ∇ θ (Loss(θ)) // Update network parameters Get θ of the generative network G through (11) Train APE-GAN through (12) End for Output: generator D…”

“…However, real continuous-variable quantum key distribution (CVQKD) [3][4][5][6] systems face security vulnerabilities due to some deviations between the theoretical assumptions and the implementation, which gives Eve the opportunity to compromise security by stealing information from legitimate parties. The eavesdroppers can employ wavelength attacks [7,8], calibration attacks [9,10], local-oscillation (LO) intensity attacks [11], homodyne-detector-blinding attacks [12], saturation attacks [13], and other attack strategies to compromise the safety of the GMCS CV-QKD. The major idea of these practical attack strategies is to use optical device flaws to deviate from the redundant noise estimates, while the essence of the corresponding typical countermeasures is to incorporate suitable real-time monitoring modules or measurement devices into the system, which depends considerably on the accuracy of the estimated excess noise.…”

A Survey of Machine Learning Assisted Continuous-Variable Quantum Key Distribution

“…Research on ML-assisted quantum attacks [32,33], attack detection and prevention [27,[34][35][36][37][38][39][40], and methods for hacking ML-based attack prevention strategies [41,42] is not included in this survey. Huang, Liu, and Zhang [43] have previously reviewed the literature in this area.…”

“…Huang, Liu, and Zhang [43] have previously reviewed the literature in this area. We note that [33,[37][38][39][40][41][42] were published after [43] so were not included in their review. Further, this survey does not include works on ML for quantum communication in general, only addressing works specifically on the application of CV-QKD.…”