Federated Learning (FL) enables devices to collaboratively train a global model without centralizing raw data. However, heterogeneous data distributions often hinder a single global model from performing optimally across diverse clients. In this paper, we introduce two personalized federated learning methods to address the aforementioned challenge. First, we introduce a gradient-fusion approach that merges global gradients aggregated via a learnable collaboration matrix with local gradients tailored to each client's data. This fusion harmonizes collective intelligence with device-specific knowledge, thereby improving personalization under data heterogeneity. Second, we propose a gradient-decoupling method that mitigates the limitations of relying solely on a collaboration matrix for balancing knowledge sharing and personalization. By decoupling the tasks and regularizing local updates through the incorporation of extracted collective intelligence, we introduce an inductive bias that significantly improves generalization. Extensive evaluations on four computed tomography and pathology imaging datasets covering a wide range of heterogeneity show that our frameworks substantially outperform existing baselines. We also provide theoretical analyses that establish performance bounds and convergence guarantees under mild assumptions.

Federated Learning (FL) enables collaborative model training across multiple edge devices without sharing raw data, yet existing FL frameworks often assume static data domains, limiting their applicability to real-world scenarios where data evolves over time. To address this, Federated Continual Learning (FCL) integrates continual learning into FL, but conventional strategies such as data replay are impractical due to privacy and storage constraints. In this paper, we propose PP-FCL, a privacy-preserving FCL framework that mitigates catastrophic forgetting without storing sensitive client data. PP-FCL employs a server-side generative model to synthesize representative samples of previously learned tasks, enhancing data diversity, preserving characteristic class features, and refining decision boundaries. On the client side, an improved contrastive incremental learning loss and a carefully designed feature distillation method decouple old and new knowledge, ensuring a balanced trade-off between plasticity and stability. As a result, PP-FCL not only enhances the model’s representational capabilities but also adapts effectively to non-stationary data distributions, maintaining robust performance in privacy-sensitive, evolving federated environments. Empirical results on CIFAR-10, CIFAR-100 and TinyImageNet demonstrate that PP-FCL outperforms state-of-the-art baselines by approximately 5–6% in average accuracy. This substantial improvement highlights PP-FCL’s effectiveness in preserving model performance under evolving conditions, ensuring robust and adaptive learning in dynamic federated environments.

In the advancement of collaborative intelligence within the Internet of Things (IoT), federated learning (FL) enables clients to collaboratively train a global model without centralizing raw data. However, the non-independent and identically distributed (non-IID) nature of data among clients often leads to divergent local training objectives, deteriorating the performance of the aggregated global model. To address this challenge, we propose AP-CFL, a novel clustered FL algorithm that incorporates affinity propagation to dynamically discover the clustering structure of clients without the need to predefine the number of clusters. Specifically, AP-CFL calculates the mean of absolute differences of pairwise cosine similarity to effectively cluster clients based on similarities in their data distributions. Knowledge sharing is enhanced by decoupling each cluster model into a globally shared encoder and a cluster-specific classifier, and the local training objectives are modified to improve the generalization capacity of the shared encoder. Additionally, a robust strategy is introduced to manage partial client participation by employing a time and data importance index, which mitigates the adverse effects of model staleness and maintains the integrity of the clustering structure. Extensive experiments on diverse real-world datasets demonstrate that AP-CFL outperforms existing FL baselines in non-IID settings, effectively improving model quality and convergence stability.

Federated learning (FL) facilitates collaborative model training in the autonomous Internet of Things (IoT) system while preserving the privacy of local data on IoT clients. Nonetheless, the inherent non-IID characteristic of local data leads to poor convergence of a global model. Moreover, the global model fails to satisfy the personalized task demands of all clients. To address the above issues, this article studies client grouping and local model aggregation in FL from two perspectives: 1) measure of client data distribution and 2) distribution similarity among clients. To this end, a novel confidence-based similarity-aware personalized FL algorithm (FedCS) for personalized autonomous IoT is proposed by developing three key innovations, namely, a public average confidence (PAC) measure, a client grouping strategy with dynamic sampling (CGDS), and a sequential aggregated weight (SAW) strategy. Specifically, the PAC measure utilizes a public data set on the server side to estimate the client’s data distribution, which promotes a fair estimate of distribution similarity among clients while minimizing privacy risks. The CGDS strategy focuses on distribution similarity among clients and approximates the client grouping problem as an auxiliary task selection problem in multitask learning. This strategy assigns a client into multiple groups and enables the valuable information from each client to circulate among multiple groups. The SAW strategy further incentivizes more similar clients within a group to share greater knowledge and generates an adaptive aggregated weight for each client within a group. A thorough experiment on CIFAR10 and two healthcare benchmarks shows that FedCS achieves a superior performance.

The Internet of Things (IoT) is penetrating many aspects of our daily life with the proliferation of artificial intelligence applications. Federated learning (FL) has emerged as a promising paradigm enabling many intelligent IoT applications; however, the transmitted model gradients or weights still encode private information, which can be exploited to launch inference attacks. One popular way is to apply local differential privacy (LDP) into FL. However, existing work does not provide a practical solution due to two issues. First, the fine-grained range difference of weights in different layers of an FL model has not been explicitly considered. Second, the accumulated privacy budget may cause a budget explosion. In this article, we propose a local differentially private scheme to train clustered FL models on heterogeneous IoT data by using adaptive clipping, weight compression, and parameter shuffling (namely, ACS-FL), aimed at mitigating the curse of dimensionality, the amount of LDP noise, and the communication overhead of IoT devices. Empirical evaluations on MNIST, fashion-MNIST, and Federated Extended MNIST demonstrate that ACS-FL achieves a superior performance in balancing the tradeoff between privacy and utility.

Federated learning (FL) enables multiple IoT devices to collaboratively train a global model without centralizing raw data. However, achieving optimal performance for each device is challenging due to the heterogeneity of IoT data. Moreover, IoT devices, constrained in computational power and energy, necessitate innovative privacy protection solutions in FL. In this paper, we propose privacy-enhanced FL methods to address data heterogeneity in IoT. Our approach includes a dual-model framework for each device, combining a shared model to facilitate inter-client communication with a personalized model. This personalized model is developed from local aggregation, utilizing layer-wise aggregation weights generated by local hypernetworks. This integration effectively balances collective intelligence with client-specific knowledge, addressing data heterogeneity for improved personalization. Additionally, we propose a Local Gradient Shielding (LGS) method to protect training data. This method conceals gradients in locally trained personalized models, preventing data reconstruction and ensuring client privacy with minimal resource consumption. Extensive evaluations on benchmark datasets, showcasing various types and degrees of data heterogeneity, demonstrate the superiority of our approaches over current baselines.

Autonomous vehicles (AVs) rely on controller area network (CAN), which ensures the communication between massive electronic control units (ECUs) and passenger safety. Although CAN is a lightweight and reliable broadcast protocol, its vulnerability has caused CAN to confront serious security threats. Adversaries and malicious organizations can impair CAN bus in a variety of ways, such as injecting malicious messages into CAN bus. These malicious messages can directly intervene the functions inside AVs. Therefore, this article proposes a novel cross-chain-based intrusion detection for CAN bus (CCID-CAN) model that uses rule-based Valid Bit index (VBIN) model for initial intrusion detection on CAN bus inside AVs, followed by the Kalman filter and Naïve Bayes model for detecting attacks missed in the VBIN, where a cross-chain mechanism implements the exchange of attack logs among connected AVs that may not trust each other so as to optimize the Naïve Bayes detector. Afterwards, a series of experiments against several types of attacks are conducted on real vehicle supported by XPeng, and the results reveal that the CCID-CAN model outperforms existing models in terms of detection performance, time overhead, and memory footprint. In addition, attack log exchange for cross-chain networks in this proposed model is of high performance in latency, memory footprint, and throughput.

The exponential growth of Internet of Things (IoT) data has pushed the boundaries of data analysis, but it has also raised concerns about the increasing commoditization of personal privacy. The intriguing problem of trading private IoT data is the focal point of this paper. We delve into three fundamental questions: Firstly, we determine the minimum privacy that a broker must purchase from data owners to achieve aggregate statistics with a fixed accuracy goal. Secondly, we propose an innovative arbitrage-free pricing framework that empowers brokers to set prices for aggregate statistics, maximizing utility while ensuring fairness. Lastly, we address the challenge of fairly compensating data owners for their privacy losses while providing economic guarantees. To achieve these objectives, we propose a data trading framework in this paper. Our framework considers ubiquitous data correlations and the potential involvement of untrusted brokers, ensuring that the total compensations remain within the brokers’ budget; each owner is guaranteed to receive non-negative revenues, and cunning data owners cannot exploit misreported privacy demands to gain extra compensations. Experiments on the MovieLens dataset demonstrated the robustness of our framework.

The novel coronavirus has been spreading worldwide and emerged as a public health crisis. As the rapid rise of infected population count, a wide variety of stringent non-pharmaceutical interventions have been taken by cities and countries around the globe, including mobility reduction, social distancing and regional lockdown. The efficacy of these interventions is hard to quantify as individuals violate policies, travel inadvertently or deliberately, and spread the virus without themselves being infected. Furthermore, the publicly available pandemic data on infectious rates and other epidemiological data are unreliable and limited, and are even underestimated. In this paper, we intend to interpret and forecast the spreading dynamics of Covid-19 and quantify the efficacy of quarantine control adopted by Wuhan, Italy, South Korea and the United States of America, employing a hybrid model of an epidemiological model and a data-driven neural network model. Furthermore, since the Covid-19 has prompted global travel restrictions, aggravated unemployment, and influenced the global economy, which exemplify the great societal cost of interventions in the battle of halting Covid-19 spreading. We intend to develop optimal quarantine control under which the tradeoff between Covid-19 containment and the societal cost of quarantine control can be optimized. Optimal quarantine control enables communities have opportunities to catch their breath to reserve healthcare resources preemptively, while the Covid-19 spreading can be halted. Our results unequivocally indicate that governments that taken stringent interventions starting from the initial stage can efficiently halt the spreading of Covid-19; furthermore, the total societal cost of such interventions is greatly smaller.

The data collected in Internet of Thing (IoT) systems (IoT data) have stimulated dramatic extension to the boundary of commercialized data statistic analysis, owing to the pervasive availability of low-cost wireless network access and off-the-shelf mobile devices. In such cases, many data consumers post their queries for urban statistic analysis in the system, like the scales of traffics, and then data contributors in IoT networks upload their contents, which can be evaluated by data brokers and responded to data consumers. However, huge volumes of devices bring large scales of data, constituting heavy burdens for data exchange. Even worse, contents in IoT systems are also sensitive as they are usually linked to private physical status of data contributors. The previous studies for IoT data trading fail to provide comprehensive estimation and pricing towards these difficulties. Therefore, this paper proposes a novel framework for the range counting trading over IoT networks by jointly considering data utility, bandwidth consumption, and privacy preservation. The range counting accumulates the number of data items falling in a concerned range of value, providing important information on the underlying data distribution. This paper first proposes a novel sampling-based method with histogram sketching for range counting estimation. The estimator is proved to be unbiased and achieves advanced performance on variance. Then the framework adopts a perturbation mechanism that can further preserve the results under differential privacy. The theoretical analysis shows that the mechanism can guarantee the privacy preservation under a given size of samples and the accuracy requirement of results. Finally, two types of pricing strategies for range counting trading are introduced for different circumstances, providing holistic consideration on how the parameters given in the estimator should be used for data trading. The framework is evaluated by estimating the air pollution levels and the traffic levels with different ranges on the 2014 CityPulse Smart City datasets. The evaluation results demonstrate that our framework can provide more accurate and reliable statistical information, with reduced bandwidth consumption and strengthened privacy preservation.

Nowadays, the social network becomes an indispensable part of people's daily life, meanwhile offers an unprecedentedly convenient access for purposive individuals to influence the opinions of network users. Current studies present a subtle persuasion approach that finds a number of key users meanwhile varies their susceptibility extent to impact the public opinion. Such persuasion is significantly critical for public security, as it could facilitate both the spreading and dispelling of malicious rumors. However, the major body of these studies enclose impractical assumptions, such that persuaders have an unlimited budget, or the costs of varying different users' susceptibilities are the same, thus rendering these works unsuitable for realistic scenarios. Therefore, this work originally proposes a more practical and generalized problem of persuasion, where varying the susceptibilities of different users holds different costs. The analysis of its non-convexity, non-submodularity and complexity shows that solving the proposed problem is nontrivial, thus inspiring us to provide an intuitive greedy algorithm. Furthermore, we design an accelerated algorithm based on the community property, which reduces the time consumption more than one order of magnitude. The acceleration is based on the intuition that the impact of a user within a proper community could be a good estimation of the impact in the whole network, while the computation of the former one is much more efficient. The relationship between two algorithms is fully analyzed, which shows the community-based algorithm can degenerate to the intuitive greedy algorithm under a specific setting. Finally, comprehensive evaluations on real-world datasets show the superiority of proposed algorithms on both effectiveness and efficiency.

The rapid growth of DNA-sequencing technologies motivates more personalized and predictive genetic-oriented services, which further attract individuals to increasingly release their genome information to learn about personalized medicines, disease predispositions, genetic compatibilities, etc. Individual genome information is notoriously privacy-sensitive and highly associated with relatives. In this paper, we present an inference attack algorithm to predict target genotypes and phenotypes based on belief propagation in factor graphs. With this algorithm, an attacker can effectively predict the target genotypes and phenotypes of target individuals based on genome information shared by individuals or their relatives, and genotype and phenotype association from genome-wide association study (GWAS). To address the privacy threats resulted from such inference attacks, we elaborate the metrics to evaluate data utility and privacy and then present a data sanitization method. We evaluate our inference attack algorithm and data sanitization method on real GWAS dataset: Age-related macular degeneration (AMD) case/control dataset. The evaluation results show that our work can effectively defense against genome threats while guaranteeing data utility.

Data privacy arises as one of the most important concerns, facing the pervasive commoditization of big data statistic analysis in Internet of Things (IoT). Current solutions are incapable to thoroughly solve the privacy issues on data pricing and guarantee the utility of statistic outputs. Therefore, this paper studies the problem of trading private statistic results for IoT data, by considering three factors. Specifically, a novel framework for trading range counting results is proposed. The framework applies a sampling-based method to generate approximated counting results, which are further perturbed for privacy concerns and then released. The results are theoretically proved to achieve unbiasedness, bounded variance, and strengthened privacy guarantee under differential privacy. Moreover, a pricing approach is proposed for the traded results, which is proved to be immune against arbitrage attacks. The framework is evaluated by estimating the air pollution levels with different ranges on 2014 CityPulse Smart City datasets. The analysis and evaluation results demonstrate that our framework greatly reduces the error of range counting approximation; and the optimal perturbation approach enables that the private counting satisfies the specified approximation degree while providing strong privacy guarantee.

The recommender system is mainly used in the e-commerce platform. With the development of the Internet, social networks and e-commerce networks have broken each other’s boundaries. Users also post information about their favorite movies or books on social networks. With the enhancement of people’s privacy awareness, the personal information of many users released publicly is limited. In the absence of items rating and knowing some user information, we propose a novel recommendation method. This method provides a list of recommendations for target attributes based on community detection and known user attributes and links. Considering the recommendation list and published user information that may be exploited by the attacker to infer other sensitive information of users and threaten users’ privacy, we propose the CDAI (Infer Attributes based on Community Detection) method, which finds a balance between utility and privacy and provides users with safer recommendations.

Releasing social network data could seriously breach user privacy. User profile and friendship relations are inherently private. Unfortunately, sensitive information may be predicted out of released data through data mining techniques. Therefore, sanitizing network data prior to release is necessary. In this paper, we explore how to launch an inference attack exploiting social networks with a mixture of non-sensitive attributes and social relationships. We map this issue to a collective classification problem and propose a collective inference model. In our model, an attacker utilizes user profile and social relationships in a collective manner to predict sensitive information of related victims in a released social network dataset. To protect against such attacks, we propose a data sanitization method collectively manipulating user profile and friendship relations. Besides sanitizing friendship relations, the proposed method can take advantages of various data-manipulating methods. We show that we can easily reduce adversary's prediction accuracy on sensitive information, while resulting in less accuracy decrease on non-sensitive information towards three social network datasets. This is the first work to employ collective methods involving various data-manipulating methods and social relationships to protect against inference attacks in social networks.

The spread of rumors in Online Social Networks (OSNs) poses great challenges to the social peace and public order. It is imperative to model propagation dynamics of rumors and develop corresponding countermeasures. Most of the existing works either overlook the heterogeneity of social networks or do not consider the cost of countermeasures. Motivated by these issues, this paper proposes a heterogeneous network based epidemic model that incorporates both the network heterogeneity and various countermeasures. Through analyzing the existence and stability of equilibrium solutions of the proposed ODE (Ordinary Differential Equation) system, the critical conditions that determine whether a rumor continuously propagates or becomes extinct are derived. Moreover, we concern about the cost of the main two types of countermeasures, i.e., Blocking rumors at influential users and spreading truth to clarify rumors. Employing the Pontryagin's maximum principle, we obtain the optimized countermeasures that ensures a rumor can become extinct at the end of an expected time period with lowest cost. Both the critical conditions and the optimized countermeasures provide a real-time decision reference to restrain the rumor spreading. Experiments based on Digg2009 dataset are conducted to evaluate the effectiveness of the proposed dynamic model and the efficiency of the optimized countermeasures.