Federated Learning (FL) enables resource-constrained nodes in edge intelligence to train a global model using local data under the coordination of a server without the risk of privacy disclosure. Secure aggregation employs security primitives to encrypt and compute local gradients, enhancing the security attributes of vanilla FL. However, server-driven FL faces communication bottlenecks and high trust risks when coordinating large-scale distributed devices, and the existing secure aggregation with input validation schemes can only verify input vectors of lengths that are powers of 2. In this work, we propose VerifyDFL, a distributed secure aggregation protocol with input validation, which enables clients to locally validate the gradients of others within the decentralized federated learning (DFL) paradigm. Specifically, we propose a distributed proof approach based on Springproofs that supports arbitrary-length input validation. Clients locally verify the L-infinity and L-2 norms of others' inputs with a zero-knowledge manner. Furthermore, we employ k-regular graphs to enhance the communication topology of DFL, which guarantees that each client can securely aggregate gradients locally even when corrupted or dropped clients participate in federated training. The security analysis and proofs ensure that VerifyDFL meets the privacy protection requirements of DFL. We conduct real benchmark experiments to show that VerifyDFL optimizes the computational cost by approximately 20% over the state-of-the-art input validation protocols. Additionally, VerifyDFL enforces L-infinity and L(2 )norm correctness verification on encrypted model gradients in edge intelligence.

With the widespread adoption of UAV technology, the physical attacks targeting UAVs have become increasingly diverse, garnering growing attention. Physical attacks pose significant threats to the security of critical hardware within UAV systems, potentially leading to severe consequences such as crashes or unauthorized hijacking. Therefore, conducting in-depth research into physical attack methods on UAV systems not only provides theoretical support and strategic guidance for designing defense measures but also facilitates the optimization and tool-based application of existing attack techniques, paving new pathways for the development of anti-UAV technologies. This review begins with a systematic decomposition and detailed introduction of UAV systems from the perspective of hardware functional structures. Subsequently, it delves into vulnerabilities of UAV systems when facing physical attacks and provides a comprehensive review of existing physical attack methods. Particular attention is given to evaluating the effectiveness, technical characteristics, strengths, and limitations of these methods. Additionally, the review explores emerging physical attack techniques and the potential security threats posed by hardware extensions of UAVs in novel application domains. Furthermore, this review proposes a quantitative risk assessment framework for UAV security, systematically evaluating various physical attack methods based on attack cost, effectiveness, and likelihood. Finally, the review discusses future research directions in the domain of physical attacks on UAV systems, emphasizing the need to enhance existing technologies to strengthen anti-UAV capabilities and highlighting the importance of developing comprehensive defense strategies against physical attacks.

Searchable encryption enables writers to upload encrypted data to an untrusted cloud server, while allowing authorized readers to perform keyword searches over the encrypted content without compromising data privacy. The multi writer/multi-reader (M/M) setting better reflects real-world de ployment requirements because it aligns with the collaborative nature of many modern applications, but introduces substantial challenges in both efficiency and security. The state-of-the art scheme, delegatable searchable encryption (DSE), achieves optimal search time and forward privacy in the M/M setting. Nevertheless, DSE provides only limited security guarantees: search tokens generated by readers reveal the update count of queried keywords, and the scheme remains susceptible to collusion attacks between users and the server. In this work, we present Archon, a novel hardware-assisted multi-user searchable encryption scheme that addresses the aforementioned security limitations while maintaining sublinear search complexity. By leveraging trusted execution environments and an updatable keyword private information retrieval protocol, Archon achieves strong collusion resistance and supports flexible user revocation in multi-user settings. We implement Archon and evaluate its performance on real-world datasets to demonstrate its practical efficiency.

Blockchain technology has considerable potential to enhance vehicle platoon management by providing robust security guarantees. Nevertheless, existing blockchain-based platoon management schemes suffer from inefficiency and poor scalability, imposing excessive computation and communication overheads on the resource-constrained vehicle side. To overcome these limitations, in this paper, we propose a Blockchain-based scheme for secure and efficient Vehicle Platoon Management (BVPM). Specifically, we introduce an enhanced Practical Byzantine Fault Tolerance algorithm with Collective Signing (CS-PBFT), which leverages a communication tree structure and Schnorr multisignatures to improve blockchain performance. In addition, we propose a distributed voting algorithm for leader election based on the Subpath Signature and Raft (SS-Raft) mechanism, where a novel subpath signature quantifies path similarity among vehicles. Security analysis demonstrates that the BVPM scheme provides robust security for platoon operations against several common attacks. Extensive simulations further reveal that, compared with the existing schemes, the BVPM scheme significantly reduces computation and communication overheads.

Vehicle platoon offers numerous benefits in terms of road safety, energy efficiency, and traffic management in urban freight delivery. Privacy preservation is critical here: location information ties to customer confidentiality and reputation guarantees platoon reliability, yet most existing platoon management schemes fail to preserve privacy while achieving vehicle location-matching. Meanwhile, traditional distance calculation methods such as Euclidean distance are unsuitable for urban road layouts, and most schemes assume member vehicles must follow to unified endpoints, a rigid constraint conflicting with the scenario's needs. In this paper, we propose a privacy-preserving platoon management scheme for flexible platoon splitting in urban freight delivery (PPFPS). In detail, the PPFPS scheme leverages location and reputation to achieve flexible platoon splitting in platoon management while preserving vehicle privacy. Specially, we design an encrypted Manhattan distance calculation method (EMC) by combining bloom filters and Paillier cryptosystem, which is tailored to the road layouts in urban environments and deployed on cloud servers. The EMC method enables privacy-preserving location matching to achieve flexible platoon splitting, and reputation is used to ensure the reliability of vehicle platoon. Furthermore, the EMC method significantly minimizes the involvement of the trusted authority by introducing cloud-assisted approaches. Theoretical analysis demonstrates that the PPFPS scheme effectively preserves privacy and defends a variety of potential attacks. Simulation evaluation confirms that the PPFPS scheme supports more functions while significantly reducing computation overheads by 66.59% to 78.72% on the TA side, and maintains communication overheads of the similar order of magnitude as the existing schemes.

With the continuous expansion of network scale and the rapid evolution of diverse network applications, the propagation scope of network attacks and security threats has been increasingly extended, and such threats emerge in an endless stream, thereby posing severe challenges to network security. As a critical defense mechanism for network security, network traffic anomaly detection enables the identification of unknown attack behaviors and contributes to establishing a secure and healthy network environment. To address the issue of feature dependency in multi-dimensional network traffic anomaly detection, this paper adopts a network traffic dimensionality reduction method based on density clustering to capture different aspects of features. The association techniques could be incorporated into association-based multi-modal feature fusion for boosting the representation capacity. Furthermore, long short-term memory networks are applied to network traffic anomaly detection, enabling the model to effectively learn the complex nonlinear relationships inherent in traffic data. Experimental results on public network traffic datasets indicate that the proposed method achieves superior performance in feature mining and representation. Comprehensive evaluation results validate the efficiency and practicability of the method, with noticeable improvements in anomaly detection accuracy.

Supervised cross-modal hashing has shown outstanding precision and efficiency in approximate nearest neighbor search (ANNS) on multimodal data. However, the existing methods still face bottlenecks in precision and privacy leakage, which are specifically manifested in the quantization errors of hash codes, insufficient utilization of semantic information, and the lack of inherent adaptability to privacy scenarios. To address these challenges, we propose CrossGuard, a novel privacy-preserving cross-modal retrieval (CMR) framework. It leverages Ternary Quantization Stratified Sampling Sparse Hashing (TARDIS) to achieve high search precision, while employing Distributed Point Function and locality-sensitive hashing to enable private retrieval with sublinear communication complexity. The ternary-sign strategy in TARDIS effectively mitigates the quantization errors in hash codes and optimizes the hash function generated from the least squares method. Furthermore, a stratified sampling strategy using K-means uniformly extracts the semantic information from each label cluster. Moreover, the expressiveness of high-dimensional sparse hashing not only improves the precision, but also makes the CMR endogenously adaptive and scalable for private ANNS. Extensive experiments demonstrate that our framework not only establishes a new state-of-the-art in public retrieval through TARDIS, but also achieves an outstanding balance among privacy, efficiency, and precision with CrossGuard for private retrieval. Specifically, CrossGuard performs a private query in 0.75 ms of private evaluation time on NUS-WIDE with 100,000 entries, while achieving 79.65% image-to-text and 84.15% text-to-image end-to-end precision. See https://github.com/ighxiy/CrossGuard-TARDIS.git for the code.

Vehicle platoon, where multiple vehicles travel in formation under a leading vehicle, enhances road capacity and reduces energy consumption. Assessing the reliability of the leading vehicle through reputation management is critical for the vehicle platoon. In the domain of reputation management, reputation updating is of paramount importance. However, many existing schemes are limited by security vulnerabilities, privacy concerns, and reliance on simplistic strategies (i.e., single reputation updating and single-dimensional evaluation), with this reliance undermining their robustness. To address these issues, this paper proposes a Privacy-Preserving Dual Reputation Updating Scheme with Multi-Dimensional Feedback Scores in Vehicle Platoon (PDRU). The PDRU scheme employs a dual reputation model, which enhances the robustness of the reputation updating and provides resilience against the single reputation attack. Meanwhile, it introduces a multi-dimensional evaluation frame work that enables following vehicles to assess the leading vehicle across multiple dimensions, with feedback scores aggregated through weighted averaging according to the importance of each dimension. Moreover, it preserves the vehicle identity privacy, feedback score privacy, leading vehicle's reputation value privacy, and following vehicle's reputation value privacy. Theoretical analysis and simulation evaluation demonstrate that the PDRU scheme achieves a strong balance between robustness, privacy, security, and efficiency, particularly in reducing computation and communication overheads on the Trusted Authority (TA) side.

Existing membership inference attacks (MIAs) based on adversarial perturbations usually have the problem that the perturbation intensity is too large to cover up the model memory characteristics of member data, resulting in a decrease in attack accuracy. To this end, this paper proposes an MIAs framework based on optimal adversarial perturbations (OAP) and designs two new attack methods on white-box and black-box attack settings. Firstly, we theoretically establish the intrinsic relationship between perturbation intensity and model memorization, and demonstrate the existence of an OAP critical point that maximizes the prediction gap between member and non-member data while preserving member-specific features. Secondly, we introduce a white-box MIA method based on OAP, termed dynamic adversarial perturbation reduction (DAPR), which adaptively adjusts perturbation intensity by evaluating the gradient sensitivity of model outputs concerning perturbation magnitude. This strategy aims to amplify the discriminative behaviors between member and non-member samples while mitigating the feature masking effect induced by overly strong perturbations. Subsequently, under the black-box setting where only model outputs are accessible, we propose hybrid gradient estimation-based adversarial perturbation reduction (HGE-APR). This method combines natural evolutionary strategies and simultaneous perturbation stochastic approximation to perform multi-scale hybrid gradient direction estimation for generating an initial perturbation. A progressive reduction strategy is then applied to iteratively identify the minimal perturbation required to induce a prediction flip, thereby precisely capturing decision-boundary behaviors that differentiate member from non-member samples. Finally, we conduct systematic experimental evaluations on benchmark datasets, including CIFAR-10/100, Tiny-ImageNet, and various model architectures. The results demonstrate that the proposed DAPR and HGE-APR methods significantly outperform existing state-of-the-art approaches in attack effectiveness, generalization, and stability.

Secure Location-Based Services (LBSs) in mobile cloud have gained widespread attention in the past decade. However, previous works mainly focus on spatial or spatial keyword query services and cannot support temporal filters simultaneously, which limits the service quality in practical applications. To address the above issue, we propose an efficient Keyword-associated Spatio-Temporal Structured Encryption (KSTSE) scheme that allows conjunctive queries according to spatio-temporal range and textual keywords on encrypted data in a mobile cloud. Specifically, we first transform geographic and temporal range queries into a unified encoding existence detection problem by combining S2 encoding and prefix encoding. Next, we build an efficient encrypted existence evaluation construction based on the circular shift and coalesce Bloom filter and symmetric hidden vector encryption. Finally, to support efficient queries on large-scale datasets, we design a hierarchical index tree structure, which can dynamically prune the search space according to the keyword and spatio-temporal range during the query process, reducing the query complexity to O(logN) Rigorous security analysis and performance evaluation show that the proposed KSTSE construction is adaptively secure under reasonable leakages and performs better than state-of-the-art schemes

The widespread use of iris-based biometric authentication has prompted concerns about the insecure storage and potential misuse of iris templates. As a remedy, many template protection schemes utilizing bloom filters have been developed to ensure that templates stored in databases are secure and inaccessible to unauthorized users. However, these schemes often suffer from limited authentication accuracy or insufficient resistance to security threats. To address these challenges, we propose a hierarchical interleaved bloom filter-based biometric template protection scheme that achieves both high authentication accuracy and strong template security. Specifically, we first introduce a reliability-aware template extraction framework, in which noise injection is incorporated during image pre-processing to enable more robust statistical estimation of feature stability. Based on the augmented samples, a dual-score reliable location selection method is designed, combining a quality estimation score and a cross-sample consistency score to extract a reliable intermediate iris template and a user-specific key. This mechanism ensures that selected features are both intrinsically high-quality and statistically stable. Furthermore, we enhance the bloom filter by integrating a permutation mechanism and a hierarchical interleaved mapping strategy to reduce structural correlation, mitigate collision effects, and improve storage efficiency and template security. In addition, a secure sketch derived from the user-specific key is stored in the database to further strengthen security. Experimental results obtained from CASIA-Iris-Lamp and CASIA-Iris-Thousand datasets demonstrate that the proposed scheme can achieve an equal error rate as low as 0.75% when three iris images are used during the enrollment stage. Besides, the ablation experiments further validate the effectiveness of the proposed reliable feature selection method and the hierarchical interleaved bloom filter design. Furthermore, the comprehensive security analysis shows that the proposed scheme effectively meets essential security properties, including irreversibility, revocability, and unlinkability.

Federated learning suffers from fairness imbalance due to gradient conflicts in non-independent and indentically distributed (Non-IID) data scenarios, with model parameter transmission risking privacy leakage. Existing fair federated learning algorithms mitigate gradient conflicts effectively but lack targeted optimization for local-global model consistency under complex heterogeneous data, leading to deviated local training from global optima, impaired fairness and limited overall performance. To address this, the FedALF algorithm and LDP-FedALF algorithm were proposed: it alleviates gradient conflicts via fairness-driven objectives and hierarchical optimization; adopts ADMM for optimization, designing dual variables and quadratic penalty terms in clients' local objective functions to enforce local-global model consistency constraints, further enhancing fairness and performance; and integrates adaptive local differential privacy into model transmission for privacy protection. Through a large number of experiments, FedALF achieved an average improvement of 10.82% in test accuracy and 19.80% in fairness performance compared to existing fair federated learning algorithms. LDP-FedALF achieved an average improvement of 7.53% in test accuracy and 15.36% in fairness performance compared to existing fair federated learning algorithms.

Crowdsensing leverages a vast number of intelligent terminals and participants to collect, transmit, and analyze data, thereby enabling the perception and processing of information. In this context, user collusion can severely compromise system performance, data accuracy, and fairness. Existing auction-based schemes designed to prevent collusion often lack transparency, enabling colluding users to exploit system rules during the auction process. This not only accelerates the degradation of the system’s ecosystem but also reduces the willingness of other users to participate. Therefore, these schemes are not only ineffective in curbing user collusion but also fail to ensure the accuracy of data. To address this issue, this paper combines the all-pay bidding mechanism with virtual points and proposes a user recruitment mechanism based on all-pay auctions that satisfies incentive compatibility. Theoretical analysis and experimental results confirm that this mechanism can maintain user participation enthusiasm while effectively improving the base and accuracy of data.

As a typical crowdsourcing service in the healthcare field, medical crowdsourcing allocates tasks to large-scale professionals via edge-cloud collaborative platforms. It provides patients with health consultations and treatment plans. Nevertheless, it is non-trivial to achieve fine-grained access control for medical tasks, fine-grained trusted identification for task requesters, and flexible task selection for medical providers. Furthermore, current access control solutions cannot prevent improper access to medical tasks throughout the full lifecycle, thus making them susceptible to historical and future data breaches. To this end, this paper proposes a privacy-preserving Attribute-Based Matchmaking Encryption scheme called MatFBMC that ensures both forward and backward secrecy. It offloads massive puncture operations to the crowdsourcing platform, thereby reducing the client key rotation overhead while ensuring forward secrecy. Meanwhile, this scheme ensures backward secrecy through the embedding of the time epoch, thereby realizing lightweight privilege revocation. Furthermore, the bi-directional matches between access/identification policies and the two parties’ attributes are employed to achieve bilateral fine-grained access control. We provide security proofs for semantic security against selective Chosen Plaintext Attack and existential unforgeability under Chosen Message Attack. Based on the Java Pairing-Based Cryptographic Library and real-world medical datasets, simulations display better computational and storage performance compared to the state-of-the-art works.

In the age of digital connectivity, vast amounts of data are openly shared across social platforms, e-commerce platforms, and other online channels, potentially leading to inadvertent exposure of sensitive information. As a crucial mechanism for protecting sensitive information, anonymization techniques for structured data have already reached a relatively mature stage. Nevertheless, unstructured textual data lacks fixed structures and involves diverse entity types, making sensitive information difficult to identify through automated methods. Meanwhile, automated anonymization of unstructured text still relies heavily on manual processing, which is costly and inefficient, and thus requires further optimization. Presently, Large Language Models (LLMs) have demonstrated remarkable capabilities in semantic understanding and reasoning, but its effectiveness in anonymizing sensitive information within unstructured text still lacks systematic research. In this paper, we employ two fine-tuning approaches, namely Prefix Tuning and Low-Rank Adaptation (LoRA) Fine-Tuning, and verify that fine-tuned LLMs still achieve excellent performance in the task of anonymizing sensitive information in texts. Furthermore, we conduct a comparative evaluation and analyze their similarities and differences with traditional Named Entity Recognition (NER) and rule-based anonymization approaches. The evaluation results indicate that the fine-tuning model achieves outstanding performance in sensitive information anonymization tasks.

Given the growing integration of smart healthcare into everyday life, ensuring the security of wearable devices has become a critical concern. Continuous authentication utilizes biometrics to verify a user's identity periodically throughout the entire session, which can provide an imperceptible additional layer of protection for security. However, existing continuous authentication schemes are not suitable for healthcare due to their shortcomings in security and practicality. To this end, we present a secure multimodal continuous authentication method based on distillation learning in this paper. Specifically, data preprocess and data transformation are designed to reduce the impact of noise in raw electrocardiogram (ECG) and photoplethysmography (PPG) data on authentication accuracy, and improve the security and robustness. Then, teacher network architectures capable of achieving unimodal continuous authentication for ECG and PPG are constructed and trained. Finally, we integrate distillation learning into the feature fusion to learn a lightweight student model from ECG and PPG teacher models to achieve user-friendly and high-accuracy multimodal continuous authentication. We evaluate the proposed method using both public datasets and real-world collected ECG and PPG data, and the experimental results indicate that our method can achieve a high authentication success rate with a low error rate, providing an effective way to improve the practicality of continuous authentication.