While differential privacy (DP) has been widely adopted to strengthen privacy guarantees in federated graph learning (FGL), its application often incurs a significant accuracy-privacy trade-off. To address this limitation, we propose PrivFGL, a framework that enhances the utility performance of differentially private FGL by mitigating noise-induced heterogeneity. Through empirical analysis, we demonstrate that the accuracy degradation in existing DP-FGL frameworks stems from amplified client heterogeneity caused by randomized perturbations during training. PrivFGL mitigates this issue via Personalized Data Transformation (PDT), which adaptively aligns the feature distributions of perturbed graph data across clien. In PrivFGL, each client utilizes a local PDT module during training to process its perturbed data. Extensive experiments on two widely-used datasets, in comparison with one heterogeneity solution method, verify the effectiveness of our method in addressing the heterogeneity problem.

The intelligent vehicle (IV) manufacturing industry is flourishing, but its large-scale adoption and application require not only ensuring driving safety and high-quality service but also addressing a multitude of data security threats. Among these, authentication and unauthorized access are significant factors hindering its implementation. To ensure the secure communication and operation of IVs, we propose an new matchmaking encryption scheme with inner product, which supports bilateral control, as well as authentication with anonymity. As a cornerstone of this article, we design a secure attribute-based anonymous credential scheme, enabling authentication. Moreover, by introducing the dual-policy of inner product, the authenticating process ensures the vehicle privacy through anonymity. We formalize the definitions of authentication and anonymity and prove that our scheme satisfies these properties. Compared with the most related works, the experimental results demonstrate that our scheme provides enhanced security while maintaining comparable computational overhead. Notably, the computational performance in encryption, verification, and decryption processes is impressive. The high efficiency of these processes makes them well-suited for resource-constrained devices in IVs.

This work presents a variable frequency phase-shift modulation quasi-steady-state observer-based control (VFPSM-QSSOBC) strategy for series parallel resonant converters (LCC) employed in grid-friendly repetitive bipolar pulse generator (RBPG), which has advantages of high efficiency and low interharmonics injection. To reduce interharmonics, constant power (CP) charging is well achieved by using quasi-steady-state observer to accurately obtain the averaged charging power. The VFPSM strategy is applied to improve the efficiency and narrow down the switching frequency range. In order to validate the feasibility of the proposed design, a 2.8 kW/1.2 kV prototype is constructed. Experimental results show that the prototype has an efficiency of 95.4% and a lower total interharmonic distortion (TIHD) under the rated operation, which proves that VFPSM-QSSOBC has advantages over conventional variable frequency modulation proportion-integration control (VFM-PIC) and frequency trajectory control (VFM-FTC) in the efficiency and the input current quality.

Currently, massive graph data, including social networks and biological proteins, is extensively utilized and contains a substantial amount of sensitive information. As cloud computing advances, graph data owners are increasingly inclined to outsource their large-scale graph data to cloud servers for diverse graph data query services. However, the seemingly limitless storage and computing capabilities present both opportunities and privacy challenges that are difficult to address. Recent research has proposed various schemes for querying outsourced graph data in an encrypted state. Unfortunately, many of these schemes fail to guarantee the correctness of query results under malicious models and may inadvertently disclose sensitive information within the graph data. Constrained shortest path (CSP) queries aim to find the shortest path between two vertices while adhering to specific threshold constraints. In this work, we propose a robust verifiable query scheme called VPCS that ensures privacy-preserving CSP queries on encrypted graphs. Our scheme achieves accurate and verifiable results while protecting the privacy of critical graph data information, with the exception of the number of vertices. Extensive experiments using real-world data sets validate the effectiveness of our scheme.

The intricate properties and relevance of graph data make it difficult to collect graph statistics privately via differential privacy (DP). Traditional centralized or local DP on graph data, face challenges like third-party threats and low data utility when collecting the clustering coefficient. In this regard, we introduce GCC-SDP, a scheme for collecting distributed Graph Clustering Coefficient with Shuffled DP (SDP). GCC-SDP gathers the local wedge lists of all edges and adjacency bit vectors through SDP and random response for calculating the noisy local triangle counts. It then collects the local degree values of all users by using Laplace mechanism, followed by estimating the global clustering coefficient of the global graph data by data collector. We provide specific steps of GCC-SDP and demonstrate through theoretical analysis that GCC-SDP conforms to various DPs, with unbiased results. Empirical experiments show that GCC-SDP performs better than existing local DP-based techniques across most accuracy metrics.

Summary Knowledge graphs contain a large amount of entity and relational data, and graph neural networks, as a class of efficient graph representation techniques based on deep learning, excel in knowledge graph modeling. However, previous neural network architectures for the most part only learn node representations and do not fully consider the heterogeneity of data. In this article, we innovatively propose a privacy attack framework based on IoT, PAFI, which is able to classify entities and relations, learn embedding representations in multi‐relational graphs, and can be applied to some existing neural network algorithms. Based on this, a fine‐grained privacy attack model, FPM, is proposed, which can perform attack operations on multiple targets, achieve selectivity of target tasks, and greatly improve the generalization ability of the attack model. In this article, the effectiveness of PAFI and FPM is demonstrated by real network datasets, and compared with previous attack methods, both of which achieve good results.