This paper proposes CFMD-i, a communication-efficient clustered federated multi-domain learning framework for intrusion anomaly detection in heterogeneous Internet-of-Things (IoT) environments. In such settings, multi-source data are typically non-independent and identically distributed (Non-IID), highly imbalanced, and distributed across resource-constrained edge devices, which poses challenges to both model robustness and training efficiency. CFMD-i addresses these issues from both modeling and system perspectives. First, a federated multi-domain optimization objective is formulated to enhance cross-domain robustness through shared representation learning and adversarial domain weighting. Second, a dynamic hierarchical clustered mechanism is introduced to group clients according to model discrepancy, measured by cosine similarity, output divergence, and intra-cluster compactness, enabling a balance between global knowledge sharing and domain-aware personalization. Third, a communication-efficient optimization scheme is developed by integrating parameter-difference transmission, adaptive leapfrog communication, and quantized error-feedback updates, thereby reducing redundant communication under heterogeneous conditions. Experiments on five intrusion detection datasets demonstrate that CFMD-i consistently outperforms representative federated baselines, including Fedavg, Dis-DAGMM, DIOT, and ZeKoC. The proposed framework consistently improves F1 score across multiple attack domains while reducing communication volume by over 95%, and remains operational under packet loss rates of up to 12%, although performance degradation becomes evident under severe communication impairment.

As the Internet of Things (IoT) and artificial intelligence (AI) technologies are rapidly evolving, smart healthcare has emerged as a transformative solution to enhance healthcare quality and optimize resource allocation. This study introduces FEI-Hi, a federated edge intelligence paradigm that integrates edge computing with federated learning (FL) to enable secure and efficient medical data processing. FEI-Hi comprises three principal layers: FL layer, which facilitates cross-device collaborative training through encrypted model updates; aggregation layer, which refines the global model by consolidating updates; and edge layer, which performs local data processing and model inference. FEI-Hi leverages distributed intelligent computation, model parameter compression, and efficient node clustering to enhance the accuracy and efficiency of medical data processing significantly. By employing Wasserstein distance for clustering and parameter selection, FEI-Hi ensures model convergence and stability. Experimental results on multiple medical datasets demonstrate a 30% improvement in the model training speed and an F1-score exceeding 90%, surpassing the state-of-the-art (SOTA) benchmarks in model parameter transfer efficiency, training speed, and accuracy.

The rapid development of intelligent transportation and autonomous driving technologies, driven by the Internet of Vehicles (IoV), faces significant challenges owing to data and system heterogeneity. These challenges stem from the multidomain nature of the IoV and threats such as data leaks and model-finding attacks, which complicate data processing and model training. To address these issues, in this study, we proposed federated multi-domain learning for IoV (FMD-IoV). FMD-IoV addresses data heterogeneity by employing clustered techniques to group similar viewpoints and multidomain machine learning to map diverse data types into a unified feature space. To address the system heterogeneity caused by diverse vehicle types, the framework introduces a similarity-based aggregation method and model weight de-regularization to enhance robustness and generalizability. Experimental results demonstrated that FMD-IoV reduced the mean square error (MSE) by 0.05 on the Synthia dataset and 0.13 on the CityScape dataset compared with the state-of-the-art methods. Moreover, it maintained or improved the MSE as the number of nodes increased, demonstrating its adaptability to complex scenarios and large-scale data. These results highlight the flexibility, resilience, and efficacy of FMD-IoV in multi-view data fusion within large-scale IoV environments.

Future sixth-generation (6G) mobile communication networks are expected to include new features such as the network data analysis function (NWDAF), which will allow network operators to integrate machine learning (ML)-based data analysis techniques into their networks. This will allow NWDAF to identify, safeguard against, and handle various types of anomalous behaviors on user devices. To this end, this study applies fair federated learning (FL) to the 3GPP standard NWDAF architecture and embeds the designed multi-task ML model to detect traffic anomalies in different types of user devices. However, there is a problem of different task demands when the same ML model is used for optimization between different tasks. Therefore, a global alternating gradient projection (AGP) technique is presented in this study. It can be applied to many tasks and utilized to solve minimization problems. The two gradient projection phases comprise each iteration of the AGP. These steps update various tasks at regular intervals, thereby providing a regularized version of the gradient to the original multi-task objective function, which results in optimal task performance. The simulation results demonstrate that the proposed multi-task ML model can simultaneously detect traffic anomalies of different types of user devices in NWDAF and outperforms state-of-the-art models in detecting multi-task anomalies in NWDAF. The experimental evaluation also implied that the designed FL applies superior anomaly detection performance in NWDAF scenarios and has lower communication overhead than that of the traditional NWDAF without affecting the ML performance.

This paper proposes multiple-error interceptive voter designs for safety-critical applications. The proposed baseline voters comprise two-stage error-filters, in which the first stage includes two parallel C-elements (CEs) and the second stage includes one CE to filter soft errors. The voters have high-speed versions, any of which is embedded with a high-speed path from its original input to its output to reduce delay. Simulation results demonstrate the soft error tolerance of the proposed voters. Moreover, compared with the triple-modular-redundancy (TMR) voter that can only tolerate single soft errors, the proposed 3-input baseline and 4-input high-speed voters can tolerate double soft errors and can reduce the area-power-delay product by 77.03% and 95.05%, respectively, due to the use of a few transistors and a high-speed path. The voters are extended to intercept N-1 soft errors, N being the number of inputs of each voter. Note that the voters are also extended to make so that they can tolerate hard/permanent errors in addition to soft errors.

Soft-errors and aging are considered as two primary factors affecting the long-term reliability of aerospace integrated circuits (ICs). As one of the key components in aerospace ICs, latches play a pivotal role to ensure desirable circuit functionality. This paper presents a single-event-upset recovery latch, namely SRBML, with bias-temperature-instability (BTI)-mitigation. By optimizing its internal structure, the latch can recover from single-event-upsets (SEUs) and reduce the stress time of transistors in feedback loops to simultaneously mitigate the impact of BTI on the latch. Simulation results demonstrate that the soft error rate increase due to BTI is reduced by roughly 34% for SRBML after BTI-mitigation. In addition, the delay of SRBML is not affected, and the area and power increase are limited compared to BTI-unmitigated latches.

Deep learning-based time series classification in 6G-supported Intelligent Transportation Systems (ITS) helps transport decision-making. Deep learning classifier training necessitates a large amount of labeled data for feature extraction. Labeling time series data in 6G-supported ITS is tough. Meta-learning can be used to train deep classifiers with limited data. However, in meta-learning, the tasks are frequently modeled by a low-complexity base learner. It is unable to use more complicate and powerful structures. The meta-learning-pretrained classifier can only perform new classification problems with the same number of classes. Most pre-training strategies do not prioritize enhancing the pre-training phase’s convergence rate and lowering the computational cost. Most research work aims to improve classification performance by increasing the complexity of the classification model. However, this raises computing costs. In this paper, we propose a one-dimensional Multi-Scale Dilated Convolution Neural Network time series classifier (MSDCNN). MSDCNN combines multi-scale CNN and dilated CNN. It can extract multi-scale characteristics from time series and reduce the complexity of the classifier. Furthermore, we propose a pre-training strategy, called Meta-transfer metric Learning using Scale function (MLS). MLS allows the classifier to gain experience from different tasks with various numbers of classes. Experiments show that MLS reduces pre-training computation costs during the pre-training phase. The pre-trained classifier, without using any fine tuning techniques, achieves the highest accuracy by comparing with the state-of-the-art methods. Finally, we present a case study of applying MSDCNN and MLS to detect road accidents in 6G-supported transportation systems.

Abstract Different types of network traffic can be treated as data originating from different domains with the same objectives of problem-solving. Previous work utilizing multi-domain machine learning has primarily assumed that data in different domains have the same distribution, which fails to effectively address the domain offset problem and may not achieve excellent performance in every domain. To address these limitations, this study proposes an attention-based bidirectional long short-term memory (Bi-LSTM) model for detecting coordinated network attacks, such as malware detection, VPN encapsulation recognition, and Trojan horse classification. To begin, HTTP traffic is modeled as a series of natural language sequences, where each request follows strict structural standards and language logic. The Bi-LSTM model is designed within the framework of multi-domain machine learning technologies to recognize anomalies of network attacks from different domains. Experiments on real HTTP traffic data sets demonstrate that the proposed model has good performance in detecting abnormal network traffic and exhibits strong generalization ability, enabling it to effectively detect different network attacks simultaneously.

Facial expression cognition technology continues to face challenges from certain perspectives despite the fact that there have been significant recent learning advances in computer vision in the areas involving posture, orientation, and viewing mode of photos or videos that affects the device performance. In particular, the current distributed machine learning schemes do not consider the privacy issue in face monitoring data. Hence, this paper proposes a new federated learning framework for unsupervised multidomain face recognition of postexercise. It is a graph AE design base to ensure multiple edge devices can cooperate with each other to ensure the optimization of the common objective function of the model to enhance the efficiency and speed of the global model. In addition, a multidomain learning loss function is proposed to share the common feature representation with other related tasks to improve domain adaptability. Adversarial learning is used to improve the recognition effect of the federated framework in each domain. The proposed scheme is validated on different multidomains expression datasets and the experimental results indicate a 19% higher F1 score than the benchmark scheme in multidomain face recognition tasks.