Data complexity-based batch sanitization method against poison in distributed learning

The security of Federated Learning(FL)/Distributed Machine Learning(DML)is gravely threatened by data poisoning attacks,which destroy the usability of the model by contaminating training samples,so such attacks are called causative availability indiscriminate attacks.Facing the problem that existing data sanitization methods are hard to apply to real-time applications due to their tedious process and heavy computations,we propose a new supervised batch detection method for poison,which can fleetly sanitize the training dataset before the local model training.We design a training dataset generation method that helps to enhance accuracy and uses data complexity features to train a detection model,which will be used in an efficient batch hierarchical detection process.Our model stockpiles knowledge about poison,which can be expanded by retraining to adapt to new attacks.Being neither attack-specific nor scenario-specific,our method is applicable to FL/DML or other online or offline scenarios.

Distributed Secure State Estimation of Multi-Agent Systems Under Homologous Sensor Attacks

This paper addresses the problem of distributed secure state estimation for multi-agent systems under homologous sensor attacks.Two types of secure Luenberger-like distributed observers are proposed to estimate the system state and attack signal simultaneously.Specifically,the proposed two observers are applicable to deal with the cases in the presence and absence of time delays during network communication.It is also shown that the proposed observers can ensure the attack estimations from different agents asymptotically converge to the same value.Sufficient conditions for guaranteeing the asymptotic convergence of the estimation errors are derived.Simulation examples are finally provided to demonstrate the effectiveness of the proposed results.

Security of the Decoy State Two-Way Quantum Key Distribution with Finite Resources

The quantum key distribution （QKD） allows two parties to share a secret key by typically making use of a one-way quantum channel. Howevery the two-way QKD has its own unique advantages, which means the two-way QKD has become a focus recently. To improve the practieM performance of the two-way QKD, we present a security analysis of a two-way QKD protocol based on the decoy method with heralded single-photon sources （HSPSs）. We make use of two approaches to calculate the yield and the quantum bit error rate of single-photon and two-photon pulses. Then we present the secret key generation rate based on the GLLP formula. The numerical simulation shows that the protocol with HSPSs has an advantage in the secure distance compared with weak coherent state sources. In addition, we present the final secret key by considering the statistical fluctuation of the yield generation rate of the LM05 protocol with finite resources and the error rate.

Security Simulation of Continuous-Variable Quantum Key Distribution over Air-to-Water Channel Using Monte Carlo Method

Considering the ocean water＇s optical attenuation and the roughness of the sea surface, we analyze the security of continuous-variable （CV） quantum key distribution （QKD） based Mr-to-water channel. The effects of the absorp- tion and scattering on the transmittance of underwater quantum channel and the maximum secure transmission distance are studied. Considering the roughness of the sea surface, we simulate the performance bounds of CV QKD with different wind speeds using the Monte Carlo method. The results show that even if the secret key rate gradually reduces as the wind speed increases, the maximum transmission distance will not be affected obviously. Compared to the works regarding short-distance underwater optical communication, our research represents a significant step towards establishing secure communication between air platform and submarine vehicle.

Proof of Security of a Semi-Device-Independent Quantum Key Distribution Protocol

Semi-device-independent quantum key distribution （SDI-QKD） has been proposed by applying the quantum dimension correlation, and the security relies on the violation of quantum dimension witness inequalities. We prove the security of the SDI-QKD protocol under the depolarization channel by considering the quantum dimension witness inequalities and minimum entropy and the specific process of the QKD protocol, combining with a four- quantum-state preparation and three measurement bases. We also provide the relationship between the dimension witness value, the error rate and the security key rate by the numerical simulation.

Intercept-resend attack on six-state quantum key distribution over collective-rotation noise channels

We investigate the effect of collective-rotation noise on the security of the six-state quantum key distribution. We study the case where the eavesdropper, Eve, performs an intercept-resend attack on the quantum communication between Alice, the sender, and Bob, the receiver. We first derive the collective-rotation noise model for the six-state protocol and then parameterize the mutual information between Alice and Eve. We then derive quantum bit error rate for three interceptresend attack scenarios. We observe that the six-state protocol is robust against intercept-resend attacks on collective rotation noise channels when the rotation angle is kept within certain bounds.

A Hierarchical P2P Model and a Data Fusion Method for Network Security Situation Awareness System

A hierarchical peer-to-peer（P2P）model and a data fusion method for network security situation awareness system are proposed to improve the efficiency of distributed security behavior monitoring network.The single point failure of data analysis nodes is avoided by this P2P model,in which a greedy data forwarding method based on node priority and link delay is devised to promote the efficiency of data analysis nodes.And the data fusion method based on repulsive theory-Dumpster/Shafer（PSORT-DS）is used to deal with the challenge of multi-source alarm information.This data fusion method debases the false alarm rate.Compared with improved Dumpster/Shafer（DS）theoretical method based on particle swarm optimization（PSO）and classical DS evidence theoretical method,the proposed model reduces false alarm rate by 3%and 7%,respectively,whereas their detection rate increases by 4%and 16%,respectively.