Long-term Time Series Forecasting is crucial across numerous critical domains, yet its accuracy remains fundamentally constrained by the receptive field bottleneck in existing models. Mainstream Transformer- and Multi-layer Perceptron (MLP)-based methods mainly rely on finite look-back windows, limiting their ability to model long-term dependencies and hurting forecasting performance. Naively extending the look-back window proves ineffective, as it not only introduces prohibitive computational complexity, but also drowns vital long-term dependencies in historical noise. To address these challenges, we propose CometNet, a novel Contextual Motif-guided Long-term Time Series Forecasting framework. CometNet first introduces a Contextual Motif Extraction module that identifies recurrent, dominant contextual motifs from complex historical sequences, providing extensive temporal dependencies far exceeding limited look-back windows; Subsequently, a Motif-guided Forecasting module is proposed, which integrates the extracted dominant motifs into forecasting. By dynamically mapping the look-back window to its relevant motifs, CometNet effectively harnesses their contextual information to strengthen long-term forecasting capability. Extensive experimental results on eight real-world datasets have demonstrated that CometNet significantly outperforms current state-of-the-art (SOTA) methods, particularly on extended forecast horizons.

Identifying and detecting anomalies in industrial Internet of Things (IIoT) systems is vital for maintaining industrial safety. In IIoT scenarios, various industrial machines operate with differing periods that overlap temporally, resulting in complex multiperiodic temporal patterns. In addition, varying production tasks and environmental conditions alter sensor dependencies, complicating the modeling of intersensor dependency topologies. Existing methods, which rely on a fixed global topologies, struggle to adapt to these complex multiperiodic temporal patterns and evolving dependency topologies, leading to low anomaly detection accuracy. To tackle these problems, we propose MAGE, a multiperiodic adaptive graph evolution guided anomaly detection framework. MAGE first segments sensor data into distinct temporal periods, then employs a dynamic graph structure learning module to model evolving dependencies. Finally, a global-local association discrepancy module is employed to enhance the anomaly detection capability. Comprehensive experiments on five real-world datasets demonstrate MAGE's superior performance compared to state-of-the-art approaches.

In the realm of industrial Internet of Things(IIoT), the data pattern is subject to change over time, necessitating the implementation of online anomaly detection to adapt to the change of data pattern. Given the multiple stages in the production process in IIoT, data at different times exhibit varying periodic characteristics. Existing training methods primarily use fixed time windows, which struggle to adapt to complex time patterns, leading to decreased accuracy in anomaly detection. Furthermore, the incremental update method which utilizes a single model cannot effectively capture changing data characteristics. This paper introduces an online anomaly detection architecture named Adaptive Time Window enabled Model Pool (ATWMP). The framework utilizes a reinforcement learning model to dynamically select the optimal time window for model update and anomaly detection. Within the model pool framework, anomaly detection is conducted based on model reliability, and model updates are performed according to concept drift in order to ensure accurate adaptation to changing data features. Comprehensive experiments conducted on eight concept-drifted datasets and IIoT datasets demonstrate the superiority of this proposed method compared with other advanced methods.

In multiaccess edge computing (MEC) systems, service migration has been extensively applied to ensure service quality by migrating services to follow mobile users. The existing migration methods mainly focus on optimizing service response latency and migration costs by predicting user’s movements. However, some malicious adversaries can learn auxiliary knowledge, i.e., users’ mobility model and service migration trajectory, and launch location inference attacks to infer user locations. This leads to serious personal security threats, like malvertising, fraud and kidnapping. In this article, we propose a location privacy-aware service migration method to against adversaries’ location inference attacks in multiuser MEC systems. First, we adopt an entropy-based location privacy metric to accurately measure user’s location privacy leakage risk. Then, we formulate the service migration progress as a joint optimization problem that minimizes service response latency and location privacy leakage risk. To cope with interuser interference, we developed a multiagent soft actor–critic (MASAC) algorithm to help users collaboratively make service migration decisions. Finally, simulations based on real-world user movement trajectories were conducted to demonstrate the superiority of the proposed method. Evaluation and analysis results showed that our proposed method can effectively protect user location privacy while maintaining a low service response latency.

Detecting anomalies in multivariate time series (MTS) data collected from industrial Internet of Things (IIoT) systems is essential for a variety of applications, including smart manufacturing. Existing methods typically learn local spatiotemporal representations from nearby time points and neighboring nodes to reconstruct or predict sensor data. However, these local representations are insufficient to model the complex nonlinear topological relationships and dynamic temporal patterns of IIoT systems, which often results in a high-false alarm rate. To address this issue, we propose a new MTS anomaly detection framework called GLAD, which is based on the global–local association discrepancy. The key concept is to detect anomalies based on the difference between the global and local spatiotemporal associations of each data sample, as the association distribution of each data sample provides a more informative description. Specifically, we introduce a Gumbel-Softmax-based graph structure learning strategy to capture the global topological connections from data. Based on the topological graph structure, we utilize a graph attention network (GAT) and transformer to extract both the global and local spatiotemporal associations of each data sample. Finally, we leverage the global–local association discrepancy to effectively detect anomalies from normal data samples. Extensive experiments on five real-world data sets demonstrate the superiority of GLAD over other state-of-the-art methods.

Teacher-student learning has emerged as a promising framework for real-time video inference on mobile devices in multi-access edge computing (MEC) networks, where heavyweight teacher models are deployed on edge servers, and lightweight student models distilled from teacher models are deployed on mobile devices. To deal with data drift and maintain the inference accuracy, the student model has to be updated periodically with the help of the teacher model through a training process. Different training configurations, such as training epochs and frozen layers, lead to different accuracy improvements with different resource requirements. However, in multi-user heterogeneous MEC networks, due to resource heterogeneity and limited computing resources of edge servers, it is quite challenging to update all the student models simultaneously to achieve high inference accuracy. To address this problem, in this paper, we propose an adaptive teacher-student framework in multi-user heterogeneous MEC networks. The key idea is to adaptively make optimal updating decisions (i.e., offloading decision and configuration selection decision) for each user, where the available resources of edge servers and network conditions are taken into account. First, we model the teacher-student collaborative video inference problem as an optimization problem with the aim of maximizing the average inference accuracy. Then, we propose an evolutionary deep reinforcement learning algorithm, CEM-MASAC, to solve this problem. Finally, trace-driven simulations employing real-world bandwidth traces demonstrate the superiority of our algorithm compared to the baseline methods.

Single-cell RNA sequencing (scRNA-seq) has emerged as a powerful tool for uncovering cellular heterogeneity. However, the high costs associated with this technique have rendered it impractical for studying large patient cohorts. We introduce ENIGMA (Deconvolution based on Regularized Matrix Completion), a method that addresses this limitation through accurately deconvoluting bulk tissue RNA-seq data into a readout with cell-type resolution by leveraging information from scRNA-seq data. By employing a matrix completion strategy, ENIGMA minimizes the distance between the mixture transcriptome obtained with bulk sequencing and a weighted combination of cell-type-specific expression. This allows the quantification of cell-type proportions and reconstruction of cell-type-specific transcriptomes. To validate its performance, ENIGMA was tested on both simulated and real datasets, including disease-related tissues, demonstrating its ability in uncovering novel biological insights.

Decentralized federated learning (DFL) is a promising technology to implement distributed machine learning in Internet of Vehicles (IoV), which enables vehicles to share and aggregate models with their neighbors in a vehicle-to-vehicle (V2V) network. However, due to the high mobility of vehicles, model sharing via V2V links may fail as the topology of the V2V network is time-varying, which greatly reduces the efficiency of model aggregating and the speed of model training. To address this problem, in this paper, we propose a graph convolution network (GCN)-based topology design method, named G-DFL, to improve the training efficiency of DFL in IoV by properly selecting a subgraph of the underlay V2V network, which is referred to as overlay network, in each round of model sharing. First, by encoding the state of vehicles, we utilize a GCN to extract the features of V2V network topology to predict the effective V2V links for model sharing. In addition, to further reduce the delay of model training, we use Christofides' Algorithm to find the Hamiltonian circuit with the least delay as the overlay network. Simulation results validate that the proposed method significantly improves the model training performance in DFL compared with the other baseline methods.

To cope with user mobility and resource constraints of the edge servers, various service migration policies have been proposed in mobile edge computing (MEC) to achieve a trade-off between user-perceived delay and the service migration cost by moving the service to the user as close as possible. However, there is a risk of user location privacy leakage if a malicious eavesdropper tracks the service migration trajectory. In this paper, we investigate service migration in MEC by taking the risk of location privacy leakage into account. More specifically, we define the total cost of the system as the combination of the migration cost, user-perceived delay and the risk of location privacy leakage. We formulate the service migration problem as a Markov decision process, and propose an efficient algorithm to find the optimal solution that minimize the long-term total cost. Finally, the simulations based on real-world taxi traces in San Francisco show that the proposed method can make service migration decisions effectively protect the location privacy of users, as well as achieves a lower total cost than other baseline methods.