Traffic State Estimation (TSE) is attracting significant attention due to its importance to the Internet of Vehicles (IoV) for various applications, such as vehicle path planning. In classic IoV, the real-time traffic data collected by road side units requires transferring to the cloud server for processing. Such a centralized manner may raise privacy leakage issues. Split Federated Learning (SFL) has emerged as one of the prevalent methods to solve these issues. However, recent studies have shown that existing SFL frameworks are vulnerable to Model Inversion (MI) attacks, leading to private raw data leakage. To this end, in this paper, we propose ShieldTSE, a privacy-enhanced SFL framework for TSE in IoV. To protect privacy and maintain utility, a variational encoder-decoder based privacy-preserving feature extraction module with adversarial learning is first proposed to generate better privacy-preserved intermediate activations with a lower-dimensional feature space. Then, a hard attention based feature selection module is designed to select partial yet crucial features from intermediate activations by removing redundant sensitive features to further reduce data privacy leakage. Experimental results demonstrate that ShieldTSE achieves superior privacy-preserving ability when against training-based and optimization-based MI attacks, with an average reconstruction mean square error improvement of 18× and 35× on METR-LA and PEMS-BAY compared to the baseline without the privacy-preserving strategy, respectively. ShieldTSE also successfully maintains better model utility compared to privacy protection baselines.

With the widespread application of blockchain technology across various industries, detecting and analyzing performance bottlenecks is crucial for evaluating and optimizing blockchain system performance. However, current research needs general performance metrics for detecting and analyzing bottlenecks. Only some studies focus on this aspect within blockchain systems. To address this, this paper first proposes 18 fine-grained performance metrics to evaluate performance across various layers of blockchain systems comprehensively. Subsequently, we introduce a generalized loosely coupled performance measurement framework to capture these metrics and construct the causal relationship between them, i.e., the mesoscopic performance structure. This approach allows for the detection and analysis of performance bottlenecks. Finally, numerous experimental results demonstrate that the causality between the relevant performance metrics disappears when the system reaches a performance bottleneck. Additionally, the framework has a performance impact of less than 15% on ChainMaker.

As an emerging learning paradigm, Federated Learning (FL) enables data owners to collaborate training a model while keeps data locally. However, classic FL methods are susceptible to model poisoning attacks and Byzantine failures. Despite several defense methods proposed to mitigate such concerns, it is challenging to balance adverse effects while allowing that each credible node contributes to the learning process. To this end, a Fair and Robust FL method is proposed for defense against model poisoning attack from malicious nodes, namely FRFL. FRFL can learn a high-quality model even if some nodes are malicious. In particular, we first classify each participant into three categories: training node, validation node, and blockchain node. Among these, blockchain nodes replace the central server in classic FL methods while enabling secure aggregation. Then, a fairness-aware role rotation method is proposed to periodically alter the sets of training and validation nodes in order to utilize the valuable information included in local datasets of credible nodes. Finally, a decentralized and adaptive aggregation mechanism cooperating with blockchain nodes is designed to detect and discard malicious nodes and produce a high-quality model. The results show the effectiveness of FRFL in enhancing model performance while defending against malicious nodes.

The mobility model in urban vehicular networks is one of the most important factors that impacts the evaluation of any vehicular networking protocols via simulations. However, to obtain a realistic mobility model in the dynamic urban environment is very challenging. Recently, several studies extract mobility models from large-scale real data sets (mostly taxi GPS data) without consideration of the statuses of taxi. In this paper, we discover two simple observations related to the taxi status via mining of real taxi traces: (1) the behavior of taxi will be influenced by the statuses, and (2) the macroscopic movement is related with different geographic features in corresponding status. Based on these two observations, a novel taxi mobility model (START) is proposed with respect to taxi status and geographic region. The simulation results illustrate that proposed mobility model has a good approximation with reality in trace samples, distribution of nodes and the contact characteristics.

Recent studies have revealed that Split Federated Learning (SFL) is vulnerable to Model Inversion (MI) attacks, where the attacker can reconstruct clients’ raw data by exploiting collected features. Though achieving results, current defenses are unsatisfactory due to the limited ability to suppress the sensitive information while preserving task-conducive information within features. Since such limited ability can be attributed to insufficient disentanglement of data-feature and feature-task dependencies, we propose a Dual Dependency Disentangling framework for SFL (D3SFL) to strengthen defense ability against MI attacks while maintaining the utility. Specifically, we first propose a variable-structure data-feature dependency decoupling module, which produces privacy-preserving features by learning input-specific sub-networks, therefore enhancing the disentanglement of data-feature dependencies to hide sensitive information. Then, we propose a stochastic feature-task dependency separating module that adopts sparse binary masks to preserve the target-task-critical features and reduce sensitive information, resulting in effective disentanglement of feature-task dependencies for lower privacy leakage and better utility maintenance. Extensive experiments on image-classification datasets (CIFAR-100 and FaceScrub) and the time-series dataset (METR-LA) show that D3SFL outperforms the comparisons, achieving remarkable defense ability against MI attacks (with up to 54×, 17×, and 18× reconstruction MSE on average, respectively) while maintaining better utility (with only 0.13% and 0.06% Accuracy drops over the standard SFL on CIFAR-100 and FaceScrub, respectively, and only a 0.03 MAE increase on METR-LA over CNFGNN). Our code is available at https://github.com/Shawn-CT/D3SFL.