Causal discovery on topological event sequences is crucial for ensuring the reliability of networks. However, existing methods struggle to capture the complex causal relationships arising from concurrent events and lack robustness to incomplete event sequences. To address these issues, we propose a resilient concurrent causal discovery method, termed RCCD, enabling robust learning of causal graphs from topological event sequences. Specifically, we first introduce an influence-aware hyperedge causal attention mechanism, which incorporates event duration into the embedding representation, aggregates concurrent event features via hyperedge causal convolution, and injects network prior knowledge to capture the complex many-to-one causal interactions. Furthermore, we design a masked-based alternating causal optimization framework, which forces the model to recover masked event types based on context through self-supervised mask reconstruction, thereby enhancing the resilience of the predictor to missing data. To validate the effectiveness of our method, we conduct extensive experiments on both simulated and real-world telecommunication network datasets. Experimental results demonstrate that the proposed method significantly outperforms existing state-of-the-art methods in both accuracy and robustness, making it more suitable for real-world telecommunication network environments.

Identifying causal relationships among alarm events is crucial for attack tracing and fault localization in telecommunication networks. However, existing methods either fail to capture complex interactions due to linear modeling, or suffer from low efficiency in causal discovery caused by complex computational processes, making them unable to meet the real-time requirements of dynamic networks. To address these problems, we propose a real-time causal discovery method for topological event sequences, termed TACD. Specifically, we first propose a topology-aware attention Hawkes process that jointly embeds event type and device information, endowing the model with the ability to capture complex spatiotemporal dependencies while preserving a linear summation structure. Furthermore, we design a two-stage soft sequential constrained optimization strategy, which utilizes the strength of graph nodes to estimate the causal order of events after gradient-based optimization, and applies directional constraints via soft masking to prune false-positive edges. To validate the effectiveness of the proposed method, we conduct extensive experiments on multiple datasets. Experimental results demonstrate that our method achieves high computational efficiency while maintaining causal discovery performance, making it well-suited for dynamic telecommunication network environments. Code will be released at https://github.com/Aquariuaa/TACD upon acceptance.

Precise trajectory tracking and intelligent tactical decision-making are essential for safe au tonomous driving in complex urban environments. However, existing approaches face fundamen tal trade-offs between adaptability, formal stability guarantees, and real-time feasibility. While end-to-end deep reinforcement learning methods have attracted increasing attention, their lim ited interpretability and lack of theoretical guarantees restrict their deployment in safety-critical applications. This paper proposes a hierarchical reinforcement learning framework that modularly inte grates perception, decision-making, and control. An attention-based LiDAR-camera fusion module provides robust state estimation under noise and adverse conditions. At the tactical level, a duel ing deep Q-network selects high-level maneuvers, while a low-level iterative learning controller with Q-learning-optimized gains (ILC-Q) ensures accurate path tracking. Unlike classical ILC methods that rely on task repetitiveness, the proposed ILC-Q dynamically adapts its learning gain and forgetting factor to address non-repetitive urban driving, formally defined by variations in reference trajectories, maneuver sequences, traffic interactions, and en vironmental conditions. A composite Lyapunov-based analysis establishes input-to-state stability under bounded disturbances. Extensive experiments in the CARLA simulator demonstrate that the proposed framework sig nificantly improves tracking accuracy, collision avoidance, and driving success rate compared to PID, MPC, and fixed-gain ILC baselines, with statistically significant results across multiple ran domized runs. The system maintains real-time performance (4.1 ms per control step) and exhibits improved generalization in zero-shot evaluations on unseen urban maps. These results highlight the potential of combining control-theoretic stability with reinforcement learning adaptability for safer autonomous driving.

In the era of time-critical applications such as healthcare, Edge Computing (EC) has emerged as a pivotal paradigm for delivering low-latency, responsive services by processing data closer to end users. However, the inherent resource constraints of edge nodes pose the challenge of maintaining adequate Quality of Service (QoS) for end users while simultaneously minimizing service delays, preventing resource overloading, and maintaining lower energy consumption at edge nodes. This paper presents a Mobility-Aware Multi-Objective (MAMO) Decision-Making framework for reliable service provisioning in dynamic edge environments. Our approach integrates an AHP-based Edge-Rank mechanism for initial server prioritization with a modified Strength Pareto Evolutionary Algorithm II (SPEA2) enhanced by Resource-Aware Fitness Assignment (RAFA) and Adaptive Crowding Distance (ACD) to fine-tune the server selection. The framework simultaneously optimizes four critical objectives: minimizing service delay violations, reducing resource-stressed faulty services, reducing service relocations, and conserving energy consumption. Extensive evaluation using real-world taxi trajectories and clinical urgency levels demonstrates significant performance improvements. Our AHP-Guided SPEA2 achieves a 15.3

Sixth-generation (6G) edge networks must orchestrate heterogeneous services across massive device populations under strict latency, energy, and privacy constraints. Existing reinforcement-learning (RL) approaches for edge orchestration either ignore semantic task content, lack awareness of network topology, or incur prohibitive communication overhead when deployed in federated settings, limiting applicability to latency-critical 6G scenarios. To address these shortcomings, we propose FedSemGNN, a hierarchical federated RL framework that jointly integrates three complementary capabilities: (i) continual semantic task embeddings regularized by elastic weight consolidation for intent-aware service placement, (ii) graph convolutional network-based topology encoding that captures spatial relationships among edge nodes for structurally coherent decisions, and (iii) a two-level proximal policy optimization architecture with priority-aware scheduling and adaptive semantic thresholds tailored to diverse 6G service classes. Comprehensive evaluation on the EdgeSimPy simulator over 1000 orchestration steps against five representative baselines demonstrates that FedSemGNN achieves 39.08 ms orchestration latency (3.3× faster than flat federation), near-perfect semantic fidelity, and 21× lower communication overhead (0.72 MB). Scalability experiments spanning seven network sizes from 6 to 1000 nodes (a 167× range) confirm that fidelity is preserved and computation time grows near-linearly. These results position FedSemGNN as a control-plane foundation for privacy-preserving, semantic-aware orchestration in large-scale 6G edge deployments.

Federated Edge Computing (FEC) has emerged as a critical paradigm for delivering low-latency, high-bandwidth services to mobile users by leveraging resources across multiple administrative domains. However, maintaining stringent Quality of Service (QoS) guarantees in such dynamic, heterogeneous environments presents significant challenges, primarily driven by user mobility, resource constraints, and fluctuating network conditions. This paper provides a comprehensive review of QoS-aware service provisioning in FEC, systematically analyzing the interplay between three core research dimensions: mobility management, context-awareness, and reliability. We critically examine state-of-the-art strategies for service deployment, mobility-aware service migration, trajectory prediction, fault handling, and multi-dimensional context integration. Our analysis reveals a fragmented research landscape, where these critical aspects are often addressed in isolation, leading to significant gaps in integrated frameworks capable of holistic optimization. We identify and synthesize these gaps, highlighting the lack of solutions that simultaneously manage resource-stressed fault scenarios, noisy trajectory data, and multi-objective QoS trade-offs. Finally, the paper outlines a forward-looking research agenda, proposing a conceptual five-layer integrated framework design to guide future research, while emphasizing the need for AI-driven orchestration, blockchain-enabled cross-domain QoS contract execution, digital twin-driven full-process optimization, generative AI integration, 6G-FEC collaborative guarantees, and sustainability-aware provisioning, as well as a critical analysis of practical engineering challenges, to enable adaptive, reliable, and efficient service provisioning in future federated edge environments.

Mobile Edge Computing (MEC) facilitates rapid data handling by processing real-time data in proximity to users. However, due to constrained resources of MEC servers and strict requirements of end users, the need for optimal server placement, and associated provisioning of composite services are required to enable better resource management of the MEC servers while enabling higher Quality of Services (QoS) to end users. In this paper, we propose a framework to enable proper provisioning of the end users’ composite services considering their mobility, while facilitating placement of MEC servers and enabling load balancing to enhance QoS. We formulated the composite service placement in MEC as a Service Coverage Problem (SCP) and solved it using our Adaptive Composite Service Deployment (ASDMD) using Multilayer Perceptron (MLP)-Based Deep Q-Network (DQN). Following that, we formulated the MEC server placement as a constraint-based P-Median problem and solved it using the Constraint-Based Knapsack Heuristic (CB-KH) algorithm. To ensure higher QoS and reduction of congested servers, we used a load balancing technique to iteratively monitor the loads of the servers and transfer the congested servers’ users to the most appropriate alternatives. Experiments and simulations based on real-world datasets showcase the significance of our model in better management of resources of MEC servers while ensuring proper processing of data and enhanced QoS for end users. The results indicate a minimum of 4.8% improvement in the reduction of composite services deployed, which improves resource usage, enhancing the user coverage by a minimum of 19.8%, to serve a greater number of users, and enabling a minimum of 12.7% reduction in the average latency, enhancing the user satisfaction, compared to the state-of-the-art methods.

This research develops a two-stage machine learning framework for predicting physicochemical properties of antiviral drugs through quantitative structure-property relationship (QSPR) modeling. We analyzed a diverse data set of 59 antiviral compounds, leveraging SMILES-based molecular descriptors to predict the first stage six topological indices: First Zagreb, Second Zagreb, ABC, Randic, Harmonic, and Forgotten. The models that provided predictions closest to the actual topological index values were employed in the second stage to estimate six physicochemical properties: molar refractivity, polar surface area, polarizability, molar volume, molecular weight, and complexity. This framework showed high predictive performance, achieving coefficient of determination of 0.9950 for molecular weight and 0.9891 for polarizability. Correlation analysis demonstrated strong relations between the topological indices and molecular properties, with the Randic index having the highest at 0.9969. A comparison with previous studies was conducted to evaluate the effectiveness of the proposed framework. This integrated pipeline provides an accurate, interpretable, and scalable framework for QSPR-based prediction of physicochemical properties in antiviral drugs.

Log anomaly detection (LAD) is essential to ensure the safe and stable operation of Cyber-physical systems. Although current LAD methods exhibit significant potential in addressing challenges posed by unstable log events and temporal sequence patterns, their limitations in detection efficiency and generalization ability present a formidable challenge when dealing with evolving systems. To construct a real-time and reliable online LAD model, we propose OMLog, a semi-supervised online meta-learning method, to effectively tackle the distribution shift issue caused by changes in log event types and frequencies. Specifically, we introduce a maximum mean discrepancy-based distribution shift detection method to identify distribution changes in unseen log sequences. Depending on the identified distribution gap, the method can automatically trigger online fine-grained detection or offline fast inference. Furthermore, we design an online learning mechanism based on meta-learning, which can effectively learn the highly repetitive patterns of log sequences in the feature space, thereby enhancing the generalization ability of the model to evolving data. Extensive experiments conducted on two publicly available log datasets, HDFS and BGL, validate the effectiveness of the OMLog approach. When trained using only normal log sequences, the proposed approach achieves the F1-Score of 93.7% and 64.9%, respectively, surpassing the performance of the state-of-the-art (SOTA) LAD methods and demonstrating superior detection efficiency.

With the rapid development of the Financial Internet of Things (FIoT), many intelligent devices have been deployed in various business scenarios. Due to the unique characteristics of these devices, they are highly vulnerable to malicious attacks, posing significant threats to the system’s stability and security. Moreover, the limited resources available in the FIoT, combined with the extensive deployment of AI algorithms, can significantly reduce overall system availability. To address the challenge of resisting malicious behaviors and attacks in the FIoT, this paper proposes a trust-based collaborative smart device selection algorithm that integrates both subjective and objective trust mechanisms with dynamic blacklists and whitelists, leveraging domain knowledge and game theory. It is essential to evaluate real-time dynamic trust levels during system execution to accurately assess device trustworthiness. A dynamic blacklist and whitelist transformation mechanism is also proposed to capture the evolving behavior of collaborative service devices and update the lists accordingly. The proposed algorithm enhances the anti-attack capabilities of smart devices in the FIoT by combining adaptive trust evaluation with blacklist and whitelist strategies. It maintains a high task success rate in both single and complex attack scenarios. Furthermore, to address the challenge of resource allocation for trusted smart devices under constrained edge resources, a coalition game-based algorithm is proposed that considers both device activity and trust levels. Experimental results demonstrate that the proposed method significantly improves task success rates and resource allocation performance compared to existing approaches.

The proliferation of Internet of Things (IoT) devices and computation-intensive applications has led to unprecedented demands on network resources and computing capabilities. This article presents multiobjective adaptive learning framework for uncrewed aerial vehicle (UAV)-assisted mobile edge computing (MOALF-UAV-MEC), a novel MOALF-UAV-MEC tailored for dynamic IoT environments. The framework integrates multiobjective reinforcement learning (MORL), model predictive control (MPC), adaptive particle swarm optimization (APSO), and Lyapunov Optimization to optimize UAV trajectories, dynamic resource allocation, and system stability. MOALF-UAV-MEC addresses critical challenges in UAV-assisted mobile edge computing (MEC), including multiobjective optimization, adaptive resource allocation, energy efficiency, scalability, and quality of service guarantees. Our approach employs a unique burst mode feature for UAVs, enabling temporary performance boosts in high-demand situations. Extensive simulations demonstrate the framework's efficiency in enhancing task completion rates, energy efficiency, and long-term system sustainability. Results show a task completion rate of 94.50%, significantly outperforming existing approaches, with an average of 1890 completed tasks per UAV and a load balancing efficiency of 96%. The framework exhibits robust adaptive behavior, achieving a 38% reduction in UAV route optimization and a 55% increase in task completion during high-load periods. This research contributes to the advancement of edge computing in IoT environments, offering a scalable and adaptive solution for deploying computational resources in areas with limited infrastructure, during temporary events, or in emergency situations.

Since the field of autonomous vehicles is developing quickly, it is becoming increasingly crucial for them to safely and effectively navigate their surroundings to avoid collisions. The primary collision avoidance algorithms currently employed by self-driving cars are examined in this thorough survey. It looks into several methods, such as sensor-based methods for precise obstacle identification, sophisticated path-planning algorithms that guarantee cars follow dependable and safe paths, and decision-making systems that allow for adaptable reactions to a range of driving situations. The survey also emphasizes how Machine Learning methods can improve the efficacy of obstacle avoidance. Combined, these techniques are necessary for enhancing the dependability and safety of autonomous driving systems, ultimately increasing public confidence in this game-changing technology.

Anomaly detection for cyber-physical systems (ADCPS) is crucial in identifying faults and potential attacks by analyzing the time series of sensor measurements and actuator states. However, current methods lack adaptation to data distribution shifts in both temporal and spatial dimensions as cyber-physical systems evolve. To tackle this issue, we propose an incremental meta-learning-based approach, namely iADCPS, which can continuously update the model through limited evolving normal samples to reconcile the distribution gap between evolving and historical time series. Specifically, We first introduce a temporal mixup strategy to align data for data-level generalization which is then combined with the one-class meta-learning approach for model-level generalization. Furthermore, we develop a non-parametric dynamic threshold to adaptively adjust the threshold based on the probability density of the abnormal scores without any anomaly supervision. We empirically evaluate the effectiveness of the iADCPS using three publicly available datasets PUMP, SWaT, and WADI. The experimental results demonstrate that our method achieves 99.0 93.1 (SOTA) ADCPS method, especially in the context of the evolving CPSs.

Time series anomaly detection is crucial for improving the security and reliability of Cyber-Physical systems (CPS). While significant progress has been made, existing methods struggle to learn discriminative representations from multivariate time series with complex interactions and noise. To address this challenge, we propose a semi-supervised anomaly detection method based on denoising-aware contrastive learning, namely SSDCL, which can achieve robust performance for CPS anomaly detection using limited supervision. Specifically, we first design a similarity combination data augmentation algorithm to handle complex interactions among continuous sensor measurements and discrete actuator states. Furthermore, we develop a denoising hierarchical contrastive loss function that mitigates data noise interference while ensuring discriminative spatio-temporal representation. To validate the effectiveness of SSDCL, we conducted empirical evaluations on three publicly available CPS time series datasets including PUMP, SWaT and WADI. The experimental results show that the proposed method achieves F1 Score of 97.5%, 93.0%, and 74.4%, respectively, outperforming the state-of-the-art (SOTA) CPS anomaly detection methods.

The spread of mobile devices and IoT sensors demands robust edge computing frameworks capable of maintaining seamless service quality for users on the move. A central challenge is intelligent service migration, which must proactively relocate user services to optimal edge servers by anticipating mobility and adapting to user-specific needs. This paper introduces a comprehensive Context-Aware Service Migration (C-Migrate) framework that synergistically integrates high-fidelity trajectory prediction with context-aware optimization to address this challenge. We propose C-Migrate framework containing two main components: First, we propose SpatioFormer, an encoder-only transformer model that forecasts user mobility with high accuracy. These predictions then inform our Hybrid Simulated Annealing with Deep Q-network refinement (SA-DQN) algorithm, which formulates server selection as a constraint-based facility location problem. Our approach uniquely incorporates a multi-dimensional context model, enabling migrations that are not only spatially efficient but also personalized to dynamic user conditions, such as in mobile healthcare scenarios. Evaluated in a large-scale urban mobility case study with extensive experiments, our framework demonstrates a significant reduction of 9–11

Industrial edge computing tasks are delay-sensitive. Service providers often offload tasks to edge servers for processing, while the neglected cloud platform can ensure stable completion in special cases. However, this cloud-edge architecture may lead to a waste of resources and unnecessary latency. In this paper, tasks are guaranteed to be processed in time ensuring moderate utilization of various resources during resource allocation. We consider tightly coupled task scheduling and corresponding heterogeneous resource allocation processes, dividing this process into two subproblems, resource allocation and task scheduling. The asynchronous decision characteristics of the cloud and the edge are considered, so we use an online algorithm to solve these problems. We use the hybrid proximal policy optimization (H-PPO) algorithm for effective resource allocation, which can make a unified decision for various continuous heterogeneous resources. For the task scheduling problem, we used the DQN algorithm. The simulation experimental results show that the proposed method can save resources better and make the delay-sensitive tasks be offloaded in time.

Due to the high flexibility and wide coverage, unmanned aerial vehicle (UAV) has always been a popular issue in the area of mobile edge computing. UAV swarms can be deployed on-site to serve user equipment (UE) and handle offloading tasks. However, it’s inefficient that the uneven distribution of tasks cause some UAVs to be assigned heavy tasks while others are idle within the swarm. To address above issue, this paper investigates a cooperative offloading problem of minimizing the total system latency and the energy consumption, subject to constraints on battery capacity and execution latency. The problem is confirmed to be a challenging mixed-integer nonconvex programming problem with Non-deterministic Polynomial feature. Therefore we propose a joint UAV dynamic role switch scheme and cooperative offloading (MARSCO) algorithm to solve it efficiently, where two sub-problems are optimized iteratively. Specifically, both of them are optimized utilizing multi-agent deep reinforcement learning (MADRL) algorithm to interact with the environment for optimization. Finally, numerical results illustrate that the proposed algorithm utilizes the system resources to significantly reduce the total system latency and energy consumption compared with the benchmark algorithms.

Wireless Industrial IoT plays a crucial role in smart factories, where many sensors are rapidly generating task requests scheduled for timely responses. Maintaining information freshness is necessary but challenging. Edge networks that combine emerging serverless feathers can enable significant improvements in development efficiency and more flexible adaptation to workloads. However, the cost of scheduling cannot be ignored. Most of the present work in serverless edge computing does not consider the impact of the age of information (AoI) and cost in task scheduling. In this paper, we consider the relationship between AoI in users and cost in service providers in practical scenarios. We model the task scheduling problem in a serverless edge computing scenario as a Markov Decision Process (MDP) and consider multi-hop forwarding task scheduling with guaranteed AoI and costs under different pressures of workloads. To solve the highly dynamic problem, we design a multi-agent deep reinforcement learning algorithm based on Proximal Policy Optimization (PPO), validate it on real datasets, and experiments show that our algorithm reduces 10% cost in low workload and up to 16% AoI in the high workload situation.

This paper studies a class of attack behavior in which adversaries assume the role of initiators, orchestrating and implementing attacks by hiring executors. We examine the dynamics of strategic attacks, modeling the initiator as an attack planner and constructing the interaction with the defender within a defender–attack planner framework. The individuals tasked with executing the attacks are identified as attackers. To ensure the attackers’ adherence to the planner’s directives, we concurrently consider the interests of each attacker by formulating a multi-objective problem. Furthermore, acknowledging the information asymmetry where defenders have incomplete knowledge of the planners’ payments and the attackers’ profiles, and recognizing the planner’s potential to exploit this for strategic deception, we develop a defender–attack planner model with deception based on signaling games. Subsequently, through the analysis of the interaction between the defender and planner, we refine the model into a tri-level programming problem. To address this, we introduce an effective decomposition algorithm leveraging genetic algorithms. Ultimately, our numerical experiments substantiate that the attack planner’s deceptive strategy indeed yield greater benefits.

As a promising solution, Vehicle Edge Computing (VEC) addresses the challenge of driverless technology to process computationally intensive tasks in a latency-sensitive manner by offloading the vehicle's computationally intensive tasks to MEC servers or the cloud. However, in situations with limited server resources, reducing service latency and improving the efficiency of service request processing remains a challenging task. To tackle this problem, we propose a joint task offloading and service caching framework aimed at minimizing the cost of unmanned vehicles. Initially, we formulate the problem as a mixed-integer nonlinear programming problem, subsequently transforming it into a solvable Partially Observable Markov Decision Process (POMDP) problem. We then design and introduce a task offloading algorithm framework based on DDQNL to address this problem. The performance of the proposed algorithm is validated through comparisons with other baseline algorithms.