Recently, the success of large models has demonstrated the importance of scaling up model sizes. However, it is difficulttodirectly train large models locally on multiple mobile devices due to their intrinsic computational constraints. To address this challenge, it becomes a crucial need to train larger global models by training small local models on devices. As a distributedlearning approach, federated learning (FL) allows multiple devices to train models locally and aggregate them to form theglobal model by sharing the updated parameters with the server, thus enabling the co-training of models. This promising feature has spurred an increasing interest in exploring the collaborative training of large models. Despite the advent of existing device-heterogeneity FL approaches, they still have limitations in fully covering the parameter space of the global model. To fill this gap, we propose a novel approach called FedBRB (Block-wise Rolling and weighted Broadcast). The core idea of FedBRB is to utilize local models of small devices to train all modules of a large global model and broadcast the trained parameters to the entire space, thereby enabling faster information sharing. This approach not only improves training efficiency but also fully utilizes limited computational resources. Experiments demonstrate that FedBRB can produce significant performance gains, achieving state-of-the-art results. Additionally, this paper provides theoretical and experimental analyses of FedBRB convergence, thereby paving a theoretical ground and providing practical guidance for further research and application of the FedBRB method.

AI painting differs from general image stylization in that it requires an explicit stroke-by-stroke process with a plausible painting order. Existing methods often struggle to preserve semantic content, maintain convincing painterly strokes, and follow human-like painting logic simultaneously. We propose LogicPainter, a human-logic-inspired neural painting framework that explicitly models foreground-first/background-second painting, coarse-to-fine layered rendering, and diffusion-driven stroke refinement. Given an input image, LogicPainter first extracts a sketch prior and salience masks, and then renders the foreground object and background in two successive stages using controllable recoloured strokes. A layer renderer progressively refines the canvas from coarse structure to fine details through differentiable rendering and feature-guided optimisation in the stroke-canvas space. Experiments on CelebA, ImageNet, and real-world photographs show that LogicPainter achieves a better balance between content fidelity and painterly appearance than representative learning-based painting methods. User studies further indicate that its painting process is perceived as closer to human painting logic, while ablation studies verify the effectiveness of the proposed two-stage design, layered renderer, and refinement mechanism.

Space–air–ground integrated networks (SAGINs) are expected to support ubiquitous communication and sensing in next-generation wireless systems. In this paper, we develop a stochastic-geometry-based analytical framework for a three-tier integrated sensing and communication (ISAC) network composed of ground base stations (BSs), unmanned aerial vehicles (UAVs), and low-earth-orbit (LEO) satellites. In the proposed architecture, ground BSs are spatially distributed according to an independent homogeneous Poisson point process (HPPP), UAVs perform sensing and uplink transmission with horizontal locations modeled by an HPPP, and LEO satellites provide relay/backhaul support with ground projections following an independent HPPP. Analytical expressions for the hop-wise communication coverage probability and the single-UAV sensing detection probability are derived. The end-to-end coverage probability is rigorously formulated as a joint event, and a tractable product-form expression is obtained by exploiting the independence of the three-tier spatial processes. Monte Carlo simulations validate the analysis and illustrate how satellite relaying, UAV deployment parameters, and propagation conditions affect communication reliability and sensing detectability. The proposed framework provides useful benchmark-level insights for the design of SAGIN-enabled ISAC systems and establishes analytical building blocks for future studies on cooperative sensing fusion.

Technological advancements in IoT-driven smart healthcare, characterized by interconnected de vices and massive outsourced data, have driven explosive data growth and increased the demand for composite queries that combine equality-range and fuzzy keyword searches. Due to limited local resources, data are increasingly outsourced to untrusted clouds, which exacerbates privacy risks. However, existing privacy-preserving mechanisms treat these query types separately, lack ing seamless integration and creating barriers to secure execution. To address these challenges, we propose CQED, a secure and efficient composite query scheme over encrypted healthcare data. At its core, CQED employs a computationally unified homomorphic inner-product framework: fuzzy keyword matching encodes keywords into uni-gram vectors and measures similarity via squared Euclidean distance, while equality-range queries are transformed into inner-product cal culations using 0-Encoding and 1-Encoding. Based on this unified design, we develop Secure Fuzzy Keyword Search (SFKS) and Secure Range Query (SRQ) algorithms to support composite queries directly over encrypted databases. In addition to query functionality, we further enhance privacy protection. Specifically, to mitigate single-dimensional privacy leakage in tree-based indexes, we introduce a Secure Retrieval Protocol (SRP) under a dual-cloud model, enabling verification of k-d tree query conditions without exposing plaintext information. In this way, CQED integrates query functionality with privacy preservation in a computationally unified framework. We pro vide formal security proofs for SFKS, SRQ, and CQED, and conduct extensive experiments on real datasets. The results demonstrate that CQED achieves both practicality and efficiency, confirming its utility for secure data outsourcing in privacy-compliant healthcare environments.

Location point set similarity range query aims to retrieve candidate point sets that are similar to the given point set in terms of location distribution patterns and geographical features, and it is vital in GIS (Geographic Information Systems), IoT (Internet of Things), and biometrics. Due to the economic and flexible advantages of cloud services, location data is frequently outsourced to cloud servers, which simultaneously increases the risk of privacy breaches. To address this, service providers choose to encrypt data before outsourcing. However, the existing schemes for similarity range query of encrypted location point sets have some problems, such as high computational complexity of measurements, which limit the query efficiency and security of schemes. To tackle these problems, this paper achieves the efficient and privacy-preserving location-point-set similarity range query for cloud computing (LPSQ). Firstly, we propose a lightweight similarity measurement called Geo-Jaccard similarity, to reduce the time complexity to $O(n)$. Secondly, to enhance the efficiency of the scheme, we integrate the kd-tree with pivot point technology to construct a pkd-tree, and design a corresponding filtering and verification algorithm. Thirdly, to enhance the security of our scheme, we encrypt the pkd-tree using a mix of matrix encryption and SHE (Symmetric Homomorphic Encryption), and design a series of protocols under SHE, such as SHE batch minimum value calculation protocol, SHE division protocol, and the approximation algorithm for computing Jaccard similarity securely. Finally, we prove that the security of our proposed LPSQ achieves CPA (Chosen Plaintext Attack) security. Furthermore, we conduct experiments to assess the performance, and the results demonstrate that LPSQ achieves sublinear search efficiency, while Geo-Jaccard similarity proves effective for similarity range queries on location point sets.

Federated Graph Learning (FGL) has emerged as a compelling paradigm for collaboratively training a global model while preserving the privacy of multi-source graphs. Nonetheless, FGL faces a critical challenge of data heterogeneity, where semantic and structural discrepancies across clients significantly degrade its performance. Although existing methods attempt to calibrate client-specific graph distributions during federated training, they inevitably fall short in aligning the optimization behaviors across clients due to dynamic parameter updates, thereby inducing a bottleneck in generalization improvement. To tackle this challenge, we propose a solution from a new perspective of prior refinement, which seeks to proactively harmonize client graph distributions before the federated training. In particular, we propose a Federated Graph Harmonization (FedGH) framework that exploits the generative strengths of graph diffusion models to perform prior refinement of local graphs. In a nutshell, FedGH designs a conditional diffusion mechanism on each client that synthesizes pseudo-graphs encapsulating both feature and structural priors, thereby facilitating explicit correction of inter-client distributional bias. On the server side, we employ the graph contrastive learning between various client-specific pseudo-graphs to incorporate the global information, subsequently guiding local data reconstruction. Importantly, model-agnostic FedGH can be seamlessly deployed as a plug-and-play module to be easily integrated with existing FGL architectures. Extensive experiments demonstrate that FedGH consistently outperforms state-of-the-art FGL baselines.

Data-free knowledge distillation-based one-shot federated learning (OSFL) trains a model in a single communication round without sharing raw data, making OSFL attractive for privacy-sensitive medical applications. However, existing methods aggregate predictions from all clients to form a global teacher. Under non-IID data, conflicting predictions dilute each other during averaging, yielding less informative soft labels that weaken distillation. We propose FedBiCross, a personalized OSFL framework with three stages: (1) clustering clients by model output similarity to form coherent sub-ensembles, (2) bi-level cross-cluster optimization that learns adaptive weights to selectively leverage beneficial cross-cluster knowledge while suppressing negative transfer, and (3) personalized distillation for client-specific adaptation. Experiments on four medical image datasets demonstrate that FedBiCross consistently outperforms state-of-the-art baselines across different non-IID degrees.

Community search, capable of discovering highly cohesive communities from large-scale graphs, has been widely researched in many fields, e.g., recommender systems and community analysis. In recent years, numerous studies have focused on privacy-preserving community search. However, existing solutions cannot balance well the requirements of structure and multi-attribute cohe siveness effectively. To address this challenge, an efficient and privacy-preserving scheme named K-Core Multi-Attribute Community Search (KMCS) is proposed based on the attribute community graph. Specifically, to improve search efficiency, on the one hand, we utilize an improved core de composition tree to index the attribute community graph. Furthermore, inspired by the Hamming distance and the K-Core inequality, we design a matrix-operation-based filtering algorithm under plaintext. On the other hand, a unified encoding mechanism is innovatively proposed, which re duces the computational cost by embedding multiple attribute vectors into the attribute matrix. In addition, based on this mechanism, we design an efficient plaintext verification algorithm. To preserve data security, firstly, using symmetric homomorphic encryption (SHE) and lightweight matrix encryption, we design a secure filtering scheme and a secure verification scheme to preserve the privacy of the structure and multi-attribute cohesiveness search. Secondly, the concrete KMCS scheme is presented, which protects outsourced data, query requests, and query results, while additionally ensuring the security of the core decomposition tree's access patterns through obfus cation techniques. The security analysis reveals that KMCS scheme can achieve all our expected security goals. Finally, through performance evaluation, extensive experiments are performed on real community network datasets, demonstrating that KMCS is efficient and practical.

Accurate classification of encrypted IoT traffic remains challenging due to evolving applications and distribution shifts. This work presents a self-updating hybrid meta-learning framework that integrates Bayesian neural networks (BNNs) for uncertainty-aware update triggering with a random forest (RF) meta-classifier for robust decision fusion. The proposed design improves scalability and interpretability through feature-importance analysis and lightweight ensemble learning. Prediction instability is quantified using the Hellinger distance, avoiding normalization overhead and enabling an adaptive familiarity score via a tunable parameter alpha. Experimental results on encrypted traffic datasets demonstrate significant gains in reliability, achieving up to 95.7% accuracy and 0.95 macro-F1, and effective selective retraining under distribution shifts.

As a paradigm for encouraging users to contribute data spontaneously, mobile crowdsensing (MCS) has received considerable attention recently. It is crucial to evaluate the truthfulness of MCS data by proper truth discovery mechanisms. Although recent truth discovery schemes can determine truthful information, they either provide limited privacy preservation or have heavy computation and communication overheads. Moreover, most of them are not resilient to malicious faults and active attacks. To tackle the above problems, we propose two fault-tolerant and privacy-preserving truth discovery solutions. Our first scheme is mainly used for scenarios with a relatively stable number of users, where participants do not frequently join or leaves. Integrating ring signature with the perturbation technique, we design an anonymous and privacy-preserving truth discovery scheme, namely RsAnonTD, which can achieve privacy preservation and resist active attacks. To address the challenge with dynamically changed workers, we devise a multi-client inner product functional encryption scheme with a lightweight zero-knowledge proof protocol (namely McFeKDeTD) for defending against active attacks. The security analysis shows that both schemes can preserve the privacy of sensory data, weights, and estimated truths while resisting active attacks, thereby guaranteeing fault tolerance. Extensive experiments demonstrate that our designs achieve superior performance than other schemes in terms of accuracy, convergence speed, and system overheads. For example, compared with the state-of-the-art approach RPTD-II, which has a security level comparable to ours, our proposed schemes, RsAnonTD and McFeKDeTD, reduce the computational overheads approximately by 98% and 69%, respectively.

Mobile Augmented Reality (MAR) apps can collect and process highly sensitive data such as spatial maps and biometrics, yet their privacy policies remain largely understudied. Prior audits of app privacy policies have typically focused on a single legal framework, such as the GDPR. Meanwhile, 20 U.S. states have comprehensive privacy laws in effect, creating a fragmented and rapidly evolving set of privacy policy obligations. To date, no study has systematically audited privacy policies against this emerging body of state-level legislation. In this paper, we present the first large-scale audit of MAR privacy policies under U.S. state privacy laws. We construct a dataset covering the MAR ecosystem, including 8,013 Google Play MAR app metadata records worldwide, and a U.S.-based subset with 6,620 APKs and 6,426 privacy policy files. We further derive an auditable disclosure taxonomy with 5 baseline requirements, 10 triggered requirements, and 4 logic chains, and build a validated four-stage automated pipeline that produces traceable, evidence-grounded disclosure judgments. Our audit reveals widespread disclosure gaps: 44.62\% of audited policies exhibit severe disclosure omissions, with each missing more than eight requirements, and four privacy-policy requirements have violation rates above 90\%. These findings suggest that MAR privacy disclosures are not keeping pace with the growing complexity of U.S. state privacy regulation. We release our dataset, taxonomy, and auditing pipeline to support future research on scalable privacy compliance auditing.

Backscatter communication (BackCom) technology holds great promise by enabling devices to send data through reflecting ambient signals. Implementing BackCom in Internet of Things (IoT) networks can effectively address energy sustainability. However, BackCom currently faces two main challenges: 1) a relatively short transmission range and low throughput, resulting from the double-path transmission of the BackCom; 2) existing performance analytical models are application-limited to IoT scenarios, because most of them consider a relatively idealized device assumption such as uniform device distribution, while the BackCom is highly sensitive to device distributions because of its environment-dependence characteristics. To tackle these challenges, we introduce directional clustered BackCom networks (DIR-CBackCom Nets) where devices are organized in clusters and equipped with directional antennas. In DIR-CBackCom Nets, the interference is unique because of the combination of directional antennas and cluster distributions. Based on DIR-CBackCom Nets, we develop a general performance analysis model, which is not only applicable to our DIR-CBackCom Nets but also adaptable to CBackCom Nets with omnidirectional antennas (OMN-CBackCom Nets) and BackCom Nets with uniform device distribution. In the performance analysis, we highlight the relationship between communication performance and key parameters related to both cluster distribution and directional antennas. Extensive simulations validate the accuracy of our model. Our analytical model can be employed to optimize network performance by adjusting antenna configuration and device distribution.

The Execute-Order-Validate (EOV) framework has been used to improve the scalability of blockchains by concurrently executing transactions. However, the EOV framework also poses a critical performance issue. Specifically, when multiple transactions access the same data, only one of them can be committed eventually while the others are aborted due to the strong concurrency control restriction. This inefficiency makes the EOV framework far from practicality since there always exist hotspot variables that can be frequently accessed in real-world scenarios, such as the Fungible Token (FT) and Non-Fungible Token (NFT) online marketplace. In this paper, we propose RapidSnail, a novel EOV framework that enables transactions to execute based on the uncommitted data to reduce the transaction abort rate in such scenarios with hotspot variables. We first propose a new read-write set representation and a concurrency execution schedule algorithm in the execution phase to maintain the concurrent efficiency. Then we propose an effect-based conflict graph construction algorithm in the order phase to handle the conflict transactions based on the new read-write set. Finally, we propose a concurrent commitment schedule algorithm to adopt the new read-write set to validate the transactions concurrently in the validation phase. Our experiment results show that RapidSnail increases the throughput by at least 4x compared to the state-of-the-art EOV framework under high contention workload. More specifically, RapidSnail reduces the abort rate by 50%, and achieves at least 4x speedup in the order phase and 2.94x speedup in the validation phase over the existing EOV frameworks.

Despite its great potential in next-generation network applications, the integrated sensing and communication (ISAC) network suffers from security challenges stemming from shared spectrum and the inherent openness of wireless channels. To tackle this emerging challenge, this paper investigates the integration of artificial noise (AN) security technology into ISAC networks. A unified analytical framework utilizing stochastic geometry is developed to analyze the trade-offs between sensing and communication performance. The framework incorporates comprehensive 3D spatial modeling essential for aerial-terrestrial integration scenarios. Closed-form expressions for the probability of detection and coverage are derived, enabling quantitative analysis of performance limits under constrained wireless resources. The proposed AN-ISAC architecture incorporates artificial noise for enhanced security, addressing privacy leakage and eavesdropping threats without complex key management. Simulation results validate the theoretical models, highlighting the architecture’s effectiveness in maintaining high security and performance standards in dynamic network environments. Analysis reveals optimal operating boundaries where artificial noise strategically balances security enhancement against sensing and communication functionality. Distance-optimized AN deployment and height-aware resource allocation demonstrate significant performance improvements over conventional approaches. This work establishes critical foundations for practical secure ISAC deployments, with future research extending to mobile aerial networks and adaptive interference management.

Federated Learning (FL) enables collaborative model training without exposing raw data; however, untrusted client updates make the system vulnerable to backdoor attacks, which inject persistent malicious behaviors while maintaining high benign accuracy. Many existing defenses are difficult to deploy in practice. Some require clean validation data or extra verification, while others rely on aggressive noise injection or strict clipping, which may reduce model utility and destabilize training, especially in Non-IID and cross-device FL settings. Meanwhile, many clustering-based defenses rely on cluster-level hard filtering, such as retaining the dominant cluster while removing or strongly down-weighting clients outside it. Such group-level decisions may mishandle borderline or minority benign clients under data heterogeneity or adaptive attacks. To address these issues, we propose SpecGate, a robust aggregation framework that models client updates through spectral decomposition. It further introduces a coordinate-wise energy difference and produces a three-dimensional feature vector consisting of residual strength, low-rank projection consistency, and parameter energy-distribution distance. We then use robust standardization and Local Outlier Factor (LOF) to generate outlier scores that reflect local neighborhood structure, and map them through temperature scaling into continuous soft weights for soft suppression rather than strict exclusion. Experiments on MNIST and CIFAR-10 under IID, Non-IID, and cross-device settings demonstrate that SpecGate consistently reduces attack success rates while maintaining strong main-task accuracy and robustness.

The Internet of Agents (IoA) is rapidly gaining prominence as a foundational architecture for interconnected intelligent systems, designed to facilitate seamless discovery, communication, and collaborative reasoning among a vast network of Artificial Intelligence (AI) agents. Powered by Large Language and Vision-Language Models, IoA enables the development of interactive, rational agents capable of complex cooperation, moving far beyond traditional isolated models. IoA involves physical entities, i.e., Wireless Agents (WAs) with limited onboard resources, which need to offload their compute-intensive agentic AI services to nearby servers. Such servers can be Mobile Agents (MAs), e.g., vehicle agents, or Fixed Agents (FAs), e.g., end-side units agents. Given their fixed geographical locations and stable connectivity, FAs can serve as reliable communication gateways and task aggregation points. This stability allows them to effectively coordinate with and offload to an Aerial Agent (AA) tier, which has an advantage not affordable for highly mobile MAs with dynamic connectivity limitations. As such, we propose a two-tier optimization approach. The first tier employs a multi-leader multi-follower Stackelberg game. In the game, MAs and FAs act as the leaders who set resource prices. WAs are the followers to determine task offloading ratios. However, when FAs become overloaded, they can further offload tasks to available aerial resources. Therefore, the second tier introduces a Double Dutch Auction model where overloaded FAs act as the buyers to request resources, and AAs serve as the sellers for resource provision. We then develop a diffusion-based Deep Reinforcement Learning algorithm to solve the model. Numerical results demonstrate the superiority of our proposed scheme in facilitating task offloading.

In an era marked by rapid advancements in connectivity and information storage technologies, prioritizing the safeguarding of data transmission is crucial. This research addresses the critical need for robust encryption methods by leveraging the inherent properties of chaotic systems. A novel chaos-based encryption system is introduced, utilizing a hybrid configuration of two 1-D chaotic maps. These maps are optimized using various evolutionary algorithms-Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Differential Evolution (DE), and Artificial Bee Colony (ABC)-to enhance the sensitivity, unpredictability, and security of the generated key sequences. The proposed method demonstrates high resistance to brute-force and statistical attacks, as confirmed by extensive simulations. Key findings include a large key space with 96.6-bit entropy, over 50% avalanche and key sensitivity rates, and 100% success in NIST randomness tests across all variants. ABC and DE emerge as the most effective optimizers in chaotic parameter tuning. Additionally, the system is shown to be computationally efficient and scalable for resource-constrained IoT environments, making it suitable for real-time secure data transmission in smart applications.

Sea state information is significant for the guide of maritime activities of autonomous vessels. The sea state estimation (SSE) model, powered by artificial intelligence (AI), has shown great effectiveness but is susceptible to malicious data attacks. These attacks can lead to significant declines in the system's performance and result in incorrect predictions about the sea state. This study introduces SecureSSE, a strategy for protecting SSE models in autonomous marine vessels from adversarial attacks. This approach incorporates three main components: 1) the multiscale feature extraction learning (MFEL) module; 2) the feature convolution aggregation learning (FCAL) module; and 3) the perturbation examples training (PET) module. The PET module is specifically crafted to create perturbation examples that are in line with unaltered data, leveraging the capabilities of both the MFEL and FCAL modules to efficiently extract and integrate detailed features from ship motion data. Our proposed SecureSSE approach is shown to significantly improve the resilience of deep learning models against potential attacks. Through experimental testing, we have validated the effectiveness of this method in enhancing SSE. Additional ablation studies highlight the critical role of each module within the SecureSSE framework. To our knowledge, this is the first study to address adversarial attacks in this context and to propose a comprehensive defense mechanism for SSE systems in autonomous marine vessels.

Federated learning (FL) allows clients to share model gradients instead of privacy-sensitive data, which can solve the issue of data silos, but lead to the problem of data privacy leakage due to the model gradient revealing the characteristics of the training data. Privacy-preserving FL based on homomorphic encryption (HE) schemes (HE-based PPFL) can properly solve the issues of participants's data privacy leakage, but they encounter some new challenges. Existing privacy-preserving federated learning (PPFL)-based single-key HE schemes face the problem that clients can obtain others' model gradients due to the shared key and PPFL-based multikey HE (MKHE) schemes face the issues of incomplete privacy protection for models and high communication overhead due to the requirement of the collaborated decryption. Moreover, existing PPFL schemes either assume the server is always honest or the verification method is unreliable and expensive. To tackle these emerging challenges in HE-based PPFL, we propose an enhancing privacy-preserving and verifiable FL scheme. Specifically, we first construct a novel MKHE algorithm that achieves single-key decryption instead of the collaborated decryption in traditional PPFL-based MKHE. Meanwhile, we design a blockchain-based public verification method for the global model by applying a vector homomorphic hash, which can properly solve the issues of unreliable and expensive global model verification of the existing global model verification methods. Formal security analysis shows that the proposed scheme can well provide complete privacy protection and guarantee the integrity of the global model. Extensive experiments demonstrate that the proposed schemes can keep high accuracy (approximate to 95%) compared with existing differential privacy-based PPFL schemes (<= 90%). Meanwhile, the proposed schemes can achieve no decryption share size (0MB) compared to existing HE-based PPFL schemes and efficient verification compared with existing model verification methods based on bilinear maps.