Cross-Platform Social Relationship Prediction (CPSRP) aims to utilize users’ data information on multiple platforms to enhance the performance of social relationship prediction, thereby promoting socio-economic development. Due to the highly sensitive nature of users’ data in terms of privacy, CPSRP typically introduces various privacy-preserving mechanisms to safeguard users’ confidential information. Although the introduction mechanism guarantees the security of the users’ private information, it tends to degrade the performance of the social relationship prediction. Additionally, existing social relationship prediction schemes overlook the interdependencies among items invoked in a user behavior sequence. For this purpose, we propose a novel privacy-preserve Federated Social Relationship Prediction with Contrastive Learning framework called FSRPCL, which is a multi-task learning framework based on vertical federated learning. Specifically, the users’ rating information is perturbed with a bounded differential privacy technology, and then the users’ sequential representation information acquired through Transformer is applied for social relationship prediction and contrastive learning. Furthermore, each client uploads their respective weight information to the server, and the server aggregates the weight information and distributes it purposes to each client for updating. Numerous experiments on real-world datasets prove that FSRPCL delivers exceptional performance in social relationship prediction and privacy preservation, and effectively minimizes the impact of privacy-preserving technology on social relationship prediction accuracy.

The Internet of Things (IoT) has grown rapidly due to artificial intelligence driven edge computing. While enabling many new functions, edge computing devices expand the vulnerability surface and have become the target of malware attacks. Moreover, attackers have used advanced techniques to evade defenses by transforming their malware into functionality-preserving variants. We systematically analyze such evasion attacks and conduct a large-scale empirical study in this paper to evaluate their impact on security. More specifically, we focus on two forms of evasion attacks: obfuscation and adversarial attacks. To the best of our knowledge, this paper is the first to investigate and contrast the two families of evasion attacks systematically. We apply 10 obfuscation attacks and 9 adversarial attacks to 2870 malware examples. The obtained findings are as follows. (1) Commercial Off-The-Shelf (COTS) malware detectors are vulnerable to evasion attacks. (2) Adversarial attacks affect COTS malware detectors slightly more effectively than obfuscated malware examples. (3) Code similarity detection approaches can be affected by obfuscated examples and are barely affected by adversarial attacks. (4) These attacks can preserve the functionality of original malware examples.

This paper proposes a cyber security strategy for cyber-physical systems (CPS) based on Q-learning under unequal cost to obtain a more efficient and low-cost cyber security defense strategy with misclassification interference. The system loss caused by strategy selection errors in the cyber security of CPS is often considered equal. However, sometimes the cost associated with different errors in strategy selection may not always be the same due to the severity of the consequences of misclassification. Therefore, unequal costs referring to the fact that different strategy selection errors may result in different levels of system losses can significantly affect the overall performance of the strategy selection process. By introducing a weight parameter that adjusts the unequal cost associated with different types of misclassification errors, a modified Q-learning algorithm is proposed to develop a defense strategy that minimizes system loss in CPS with misclassification interference, and the objective of the algorithm is shifted towards minimizing the overall cost. Finally, simulations are conducted to compare the proposed approach with the standard Q-learning based cyber security strategy method, which assumes equal costs for all types of misclassification errors. The results demonstrate the effectiveness and feasibility of the proposed research.