As an emerging distributed learning paradigm, Federated Learning (FL) facilitates collaborative training among multiple clients without sharing raw data. However, the classic FL still faces significant challenges due to feature/model heterogeneity and catastrophic forgetting, which seriously hinder knowledge transfer and cause the forgetting of previous knowledge. To address these important challenges, we propose FBCL, a novel generalizable heterogeneity-aware Federated features and Basic-matrix Consistency Learning to balance intra-domain discriminability and inter-domain generalization. For feature/model heterogeneity, we align the similarity of feature distribution and construct the high-dimensional basic matrix with irrelevant unlabeled data, thereby overcoming communication barriers and learning generalizable representations while maintaining strict privacy preservation. For catastrophic forgetting during local updating, we introduce constraints in high-dimensional features to retain inter-domain knowledge and then extract accurate knowledge by distilling old models to preserve worthy historical information. Using real-world unlabeled public datasets, extensive experiments validate the superiority of the proposed FBCL, which outperforms the state-of-the-art methods on different scenarios of image classification.

Unmanned aerial vehicle (UAV) aided Mobile Edge Computing (MEC) has emerged as a promising technique to offer computing support for high-mobility and high-demand mobile devices (MDs). However, due to the dynamic scale of UAVs and MDs as well as their changeable resource availability and demands, it is very challenging to quickly make suitable UAV deployment plans for satisfying the real-time requirements. Existing solutions commonly adopt the centralized decision-making manner based on the global information, which leads to poor scalability, excessive search time, and repeated training costs. To address these important challenges, we propose a novel Federated deep Reinforcement learning based UAV Deployment optimization method (FRUD) for multi-UAV-aided MEC systems, aiming to minimize the average task response time via optimizing the real-time deployment locations of large-scale UAVs. In FRUD, each UAV independently conducts the deployment decision-making based on the local information of runtime environments rather than using the global information. Next, through feedback control and multi-UAV cooperation, an effective UAV deployment plan can be gradually formed. Simulation results show that the proposed FRUD well handles the UAV deployment problem in large-scale and dynamic multi-UAV-aided MEC systems and outperforms the state-of-art methods.

Multimodal sentiment analysis (MSA) is an important research topic for understanding human behavior by fusing information across modalities. Attention mechanisms in multimodal fusion highlight key information and improve recognition performance. However, conventional attention-based fusion methods effectively capture salient sentiment information while potentially overlooking subtle cues. Moreover, noise may arise both before and after multimodal fusion, which can readily disrupt downstream tasks. In this article, we leverage neighborhood information to analyze individual samples, extract contextual cues within neighborhoods, and detect residual noise after multimodal fusion. Accordingly, we propose N-AFD, a method that jointly performs alignment, fusion, and denoising. First, we present a cross-modal alignment method that constructs high-purity neighborhoods in the text representation space and aligns other modalities to text. Then, we employ a cross-modal fusion method based on neighborhood attention to capture subtle signals while preserving dominant semantics, and further extend it to a multispace and multiscale variant to enhance representational capacity. Next, we adopt an outlier-aware denoising method that masks isolated samples to reduce their interference with subsequent learning processes. Finally, extensive experiments on two benchmark datasets validate the effectiveness of the proposed N-AFD for the MSA task and demonstrate its competitive performance against other baseline methods.

Federated learning is a collaborative training paradigm designed to protect private data and is widely used in the cooperative training of Internet of Things (IoT) devices. However, despite its focus on privacy protection, federated learning remains susceptible to poisoning attacks from malicious clients. These attacks can degrade system performance and potentially lead to data privacy breaches. Moreover, real-world IoT datasets are often heterogeneous, further increasing the difficulty of detecting malicious clients. Existing defense mechanisms often struggle to effectively identify malicious clients while maintaining high model performance. To address this issue, we propose a defense mechanism called Malicious client exclusion aggregation (MaEA). This method utilizes KL divergence to preliminarily filter out anomalous clients, aggregates the remaining (preliminarily filtered) clients to obtain a pre-center model, and then identifies and excludes malicious clients by measuring their deviations from this pre-center model. We executed a series of extensive experiments on the CIFAR-10 dataset to demonstrate the effectiveness of MaEA. The results demonstrate that our approach can efficiently detect and identify malicious clients while correcting model performance.

The emerging Federated Graph Learning (FGL) offers promising collaborative training on distributed graph data. However, malicious actors may contaminate data streams by falsifying node relationships on clients or conduct adversarial attacks on edge servers, causing degraded inference and privacy leakage. Although some studies focus on privacy-protection FGL, they do not consider robustness and membership privacy amidst data pollution and adversarial attacks. Moreover, classic FGL commonly adopts FedAvg but neglects the impact of uneven information flow from distinct subtopologies. To address these important challenges, we propose GuardFGL, a novel similarity-driven FGL that extracts minimal-sufficient information from polluted data to maintain strong adversarial robustness and protect membership privacy. First, we incorporate structural-aware and feature-selection learning to explore target-relevant edges and features, avoiding privacy leakage from raw data. Next, we design an original Federated Graph Information Bottleneck (FGIB) principle to supervise extracting well-compressed information, mitigating the interference of polluted data streams. Finally, we develop a similarity-driven federated aggregation with auxiliary local information to alleviate the impact of uneven information flow. Using the real-world testbed and benchmark graph datasets, extensive experiments demonstrate that GuardFGL can achieve superior robust prediction and better protect membership privacy than state-of-the-art methods under adversarial attacks.

In Internet of Vehicles (IoV), unmanned aerial vehicles (UAVs) assisted mobile edge computing (MEC) can improve the system performance and communication range of intelligent transportation systems (ITSs). However, the resource allocation and computation offloading in UAVs-assisted IoV systems still face huge challenges due to the growing number of vehicle terminals (VTs), potential privacy leakage, and inefficient problem-solving. Existing solutions cannot adapt to such dynamic multi-UAV scenarios and meet the real-time requirements of VTs. To address these challenges, we propose RACOMU, a novel resource allocation and collaborative offloading framework for multi-UAV-assisted IoV. First, we introduce the convex optimization theory to decouple the original problem and then obtain the near-optimal allocation of transmission power and computing resources by solving the Karush-Kuhn–Tucker (KKT) condition. Next, we design a new collaborative offloading strategy with federated deep reinforcement learning (FDRL), where the offloading requests from VTs are processed in a distributed manner to approach the global optimum while preserving data privacy. Extensive experiments verify the effectiveness of the proposed RACOMU. Compared to benchmark methods, RACOMU achieves better performance in terms of task processing latency, decision-making time, and load balancing degree under various scenarios.

By integrating subgraph learning with federated learning, federated subgraph learning realizes collaborative learning of subgraph information among distributed Unmanned Agents (UAs) while protecting data privacy, offering a promising solution for graph modeling in Internet of Unmanned Agents (IUA). However, due to the various manners of collecting data on different UAs, graph data exhibits the features of Non-Independent and Identically Distributed (Non-IID), while the structures and features of local graph data on UAs are quite diverse. These factors lead to convergence difficulties and insufficient generalization ability of federated subgraph learning during the training process. To address these important challenges, we propose PFedSL, a novel knowledge-sharing Personalized Federated Subgraph Learning framework for IUA. First, a new personalized model aggregation is performed based on the confidence score of UAs and their similarity to reduce the interference of Non-IID data on model performance. Next, a parameter selective activation is introduced for model updating to handle the heterogeneity issue of subgraph structural features. Finally, an original personalized single-view contrastive learning is designed to optimize node embedding, thereby enhancing local representation consistency. Using real-world benchmark graph datasets, extensive experiments demonstrate the superiority of the proposed PFedSL. The results show that PFedSL achieves higher node classification accuracy than state-of-the-art methods in different scenarios. Meanwhile, the effectiveness of the core components in PFedSL is validated via ablation studies.

Embodied AI has notably advanced the autonomy of physical agents such as robots, vehicles, and AAVs, expanding their application scope. However, existing systems are predominantly data-driven, relying on static programming and pre-trained models. This limits their adaptability to dynamic and unforeseen scenarios. Additionally, the high computational cost of training large-scale models locally hinders their practical deployment. One promising solution lies in integrating Large Language Models (LLMs) into Embodied AI frameworks. Although LLMs excel in reasoning, coding, and perception, most existing frameworks adopt a single-LLM architecture, which restricts their effectiveness in addressing complex, multimodal tasks. The diverse strengths of individual LLMs, ranging from natural language understanding, visual processing to code generation, are seldom utilized in a collaborative and structured manner. To address these challenges, we propose a knowledge-driven framework, called EUF, that incorporates multi-LLMs into the Embodied AI architecture for AAV-as-a-Service in Intelligent Transportation Systems. Each LLM is dedicated to a specific stage of the AAV task, including user intent interpretation, adaptive path planning with code generation, and error-feedback mechanisms. Our research explores both One-shot and Segmented Code Generation approaches using various LLM-driven models to identify the optimal strategy. We conduct an in-depth analysis of different code errors to evaluate the strengths and limitations of each approach, including the feedback capabilities of the LLM-driven models. Experiments conducted in the AirSim environment demonstrate the framework’s robustness and accuracy in complex AAV path planning tasks, highlighting its practical potential for real-world ITS deployments.

Mobile Edge Computing (MEC) offers low-latency and high-bandwidth support for Internet-of-Vehicles (IoV) applications. However, due to high vehicle mobility and finite communication coverage of base stations, it is hard to maintain uninterrupted and high-quality services without proper service migration among MEC servers. Existing solutions commonly rely on prior knowledge and rarely consider efficient resource allocation during the service migration process, making it hard to reach optimal performance in dynamic IoV environments. To address these important challenges, we proposeSR-CL, a novel mobility-aware seamless Service migration and Resource allocation framework via Convex-optimization-enabled deep reinforcement Learning in multi-edge IoV systems. First, we decouple the Mixed Integer Nonlinear Programming (MINLP) problem of service migration and resource allocation into two sub-problems. Next, we design a new actor-critic-based asynchronous-update deep reinforcement learning method to handle service migration, where the delayed-update actor makes migration decisions and the one-step-update critic evaluates the decisions to guide the policy update. Notably, we theoretically derive the optimal resource allocation with convex optimization for each MEC server, thereby further improving system performance. Using the real-world datasets of vehicle trajectories and testbed, extensive experiments are conducted to verify the effectiveness of the proposedSR-CL. Compared to benchmark methods, theSR-CLachieves superior convergence and delay performance under various scenarios.

The emerging load prediction techniques support up-front and rational resource provisioning in edge systems to enhance system efficiency and Quality-of-Service (QoS). Classic prediction methods may handle loads with apparent trends, but they cannot achieve accurate prediction for highly-variable edge loads. With the advantage of sequential data analysis, recurrent neural networks (RNNs) are often used for load prediction but reveal limited generalization ability and low training efficiency. Moreover, it is hard to obtain a well-performed prediction model by discrete single-edge training with insufficient historical data. To address these important challenges, we propose a novel Multi-edge Cooperative universal framework for load Prediction with Personalized Federated deep learning (MC-2PF), enabling multi-edge cooperative training of load prediction models. Specifically, to solve the client-drift issue in federated learning (FL) caused by distinct data distribution, we customize personalized models for each edge by independent control parameters and theoretically analyze the model convergence improvement. Meanwhile, we prove the generalization bound of the MC-2PF and its universality to RNN-based prediction models through a practical example. Using the real-world testbed and load datasets, extensive experiments verify the effectiveness and practicality of the MC-2PF for different RNN-based prediction models. Compared to state-of-the-art frameworks, the MC-2PF achieves higher prediction accuracy, faster convergence, and stronger adaptiveness.

Mobile Edge Computing (MEC) distributes resources such as computing, storage, and bandwidth to the side close to users, which can provide low-latency services to in-vehicle users, thus promising a more efficient and safer driving environment. However, due to the dynamic scale of vehicle and the variability of resource requirements, it is a significant challenge to quickly obtain effective task offloading in large-scale vehicle scenarios. The existing studies generally adopt the centralized decision-making method, with long decision-making time and high computational overhead, which cannot effectively achieve good offloading decisions in large-scale scenarios. To address these problems, we propose a Multi-agent Collaborative Method for vehicular task offloading using Federated Deep Reinforcement Learning called MCM-FDRL. First, each vehicle as an agent, independently makes offloading decisions based on local information. Next, the offloading decision model of each vehicle is obtained through federated reinforcement learning training. At runtime, an effective vehicle offloading plan can be gradually developed through multi-agent collaboration. Using two real-world datasets, experiments show that the MCM-FDRL has good adaptability and scalability. Moreover, compared to the state-of-the-art methods, the task's average response time of the MCM-FDRL is reduced by 9.75%-64.90%, respectively.

Graph convolutional network (GCN) has gained widespread attention in semisupervised classification tasks. Recent studies show that GCN-based methods have achieved decent performance in numerous fields. However, most of the existing methods generally adopted a fixed graph that cannot dynamically capture both local and global relationships. This is because the hidden and important relationships may not be directed exhibited in the fixed structure, causing the degraded performance of semisupervised classification tasks. Moreover, the missing and noisy data yielded by the fixed graph may result in wrong connections, thereby disturbing the representation learning process. To cope with these issues, this article proposes a learnable GCN-based framework, aiming to obtain the optimal graph structures by jointly integrating graph learning and feature propagation in a unified network. Besides, to capture the optimal graph representations, this article designs dual-GCN-based meta-channels to simultaneously explore local and global relations during the training process. To minimize the interference of the noisy data, a semisupervised graph information bottleneck (SGIB) is introduced to conduct the graph structural learning (GSL) for acquiring the minimal sufficient representations. Concretely, SGIB aims to maximize the mutual information of both the same and different meta-channels by designing the constraints between them, thereby improving the node classification performance in the downstream tasks. Extensive experimental results on real-world datasets demonstrate the robustness of the proposed model, which outperforms state-of-the-art methods with fixed-structure graphs.

As a key technique for future networks, the performance of emerging multi-edge caching is often limited by inefficient collaboration among edge nodes and improper resource configuration. Meanwhile, achieving optimal cache hit rates poses substantive challenges without effectively capturing the potential relations between discrete user features and diverse content libraries. These challenges become further sophisticated when caching schemes are exposed to adversarial attacks that seriously impair cache performance. To address these challenges, we introduce RoCoCache, a resilient collaborative caching framework that uniquely integrates robust federated deep learning with proactive caching strategies, enhancing performance under adversarial conditions. First, we design a novel partitioning mechanism for multi-dimensional cache space, enabling precise content recommendations in user classification intervals. Next, we develop a new Discrete-Categorical Variational Auto-Encoder (DC-VAE) to accurately predict content popularity by overcoming posterior collapse. Finally, we create an original training mode and proactive cache replacement strategy based on robust federated deep learning. Notably, the residual-based detection for adversarial model updates and similarity-based federated aggregation are integrated to avoid the model destruction caused by adversarial updates, which enables the proactive cache replacement adapting to optimized cache resources and thus enhances cache performance. Using the real-world testbed and datasets, extensive experiments verify that the RoCoCache achieves higher cache hit rates and efficiency than state-of-the-art methods while ensuring better robustness. Moreover, we validate the effectiveness of the components designed in RoCoCache for improving cache performance via ablation studies.

Computation offloading utilizes powerful cloud and edge resources to process workflow applications offloaded from Mobile Devices (MDs), effectively alleviating the resource constraints of MDs. In end-edge-cloud environments, workflow applications typically exhibit complex task dependencies. Meanwhile, parallel tasks from multi-MDs result in an expansive solution space for offloading decisions. Therefore, determining optimal offloading plans for highly dynamic and complex end-edge-cloud environments presents significant challenges. The existing studies on offloading tasks for multi-MD workflows often adopt centralized decision-making methods, which suffer from prolonged decision time, high computational overhead, and inability to identify suitable offloading plans in large-scale scenarios. To address these challenges, we propose a Multi-agent Collaborative method for Workflow Task offloading in end-edge-cloud environments with the Actor-Critic algorithm called MCWT-AC. First, each MD is modeled as an agent and independently makes offloading decisions based on local information. Next, each MD's workflow task offloading decision model is obtained through the Actor-Critic algorithm. At runtime, an effective workflow task offloading plan can be gradually developed through multi-agent collaboration. Extensive simulation results demonstrate that the MCWT-AC exhibits superior adaptability and scalability. Moreover, the MCWT-AC outperforms the state-of-art methods and can quickly achieve optimal/near-optimal performance.

Federated Graph Learning (FGL) has garnered widespread attention by enabling collaborative training on multiple clients for semi-supervised classification tasks. However, most existing FGL studies do not well consider the missing inter-client topology information in real-world scenarios, causing insufficient feature aggregation of multi-hop neighbor clients during model training. Moreover, the classic FGL commonly adopts the FedAvg but neglects the high training costs when the number of clients expands, resulting in the overload of a single edge server. To address these important challenges, we propose a novel FGL framework, named SpreadFGL, to promote the information flow in edge-client collaboration and extract more generalized potential relationships between clients. In SpreadFGL, an adaptive graph imputation generator incorporated with a versatile assessor is first designed to exploit the potential links between subgraphs, without sharing raw data. Next, a new negative sampling mechanism is developed to make SpreadFGL concentrate on more refined information in downstream tasks. To facilitate load balancing at the edge layer, SpreadFGL follows a distributed training manner that enables fast model convergence. Using real-world testbed and benchmark graph datasets, extensive experiments demonstrate the effectiveness of the proposed SpreadFGL. The results show that SpreadFGL achieves higher accuracy and faster convergence against state-of-the-art algorithms.

The emerging Graph Convolutional Networks (GCNs) have attracted widespread attention in graph learning, due to their good ability of aggregating the information between higher-order neighbors. However, real-world graph data contains high noise and redundancy, making it hard for GCNs to accurately depict the complete relationships between nodes, which seriously degrades the quality of graph representations. Moreover, existing studies commonly ignore the distribution difference between feature and semantic spaces in graphs, causing inferior model generalization. To address these challenges, we propose DIB-RGCN, a novel robust GCN framework, to explore the optimal graph representation with the guidance of the well-designed dual information bottleneck principle. First, we analyze the reasons for distribution differences and theoretically prove that minimal sufficient representations in specific spaces cannot promise optimal performance for downstream tasks. Next, we design new dual channels to regularize feature and semantic spaces, eliminating the sharing of task-irrelevant information between spaces. Different from existing denoising algorithms that adopt a random dropping manner, we innovatively replace potential noisy features and edges with local neighboring representations. This design lowers edge-specific coefficient assignment, alleviating the interference of original representations while retaining graph structures. Further, we maximize the sharing of task-relevant information between feature and semantic spaces to alleviate the difference between them. Using real-world datasets, extensive experiments demonstrate the robustness of the proposed DIB-RGCN, which outperforms state-of-the-art methods on classification tasks.

In Mobile Edge Computing (MEC) systems, Unmanned Aerial Vehicles (UAVs) facilitate Edge Service Providers (ESPs) offering flexible resource provisioning with broader communication coverage and thus improving the Quality-of-Service (QoS). However, dynamic system states and various traffic patterns seriously hinder efficient cooperation among UAVs. Existing solutions commonly rely on prior system knowledge or complex neural network models, lacking adaptability and causing excessive overheads. To address these critical challenges, we propose the DisOff, a novel profit-aware cooperative offloading framework in UAV-enabled MEC with lightweight Deep Reinforcement Learning (DRL). First, we design an improved DRL with twin critic-networks and delay mechanism, which solves the Q-value overestimation and high variance and thus approximates the optimal UAV cooperative offloading and resource allocation. Next, we develop a new multi-teacher distillation mechanism for the proposed DRL model, where the policies of multiple UAVs are integrated into one DRL agent, compressing the model size while maintaining superior performance. Using the real-world datasets of user traffic, extensive experiments are conducted to validate the effectiveness of the proposed DisOff. Compared to benchmark methods, the DisOff enhances ESP profits while reducing the DRL model size and training costs.

With flexible mobility and broad communication coverage, Unmanned Aerial Vehicles (UAVs) have become an important extension of Multi-access Edge Computing (MEC) systems, exhibiting great potential for improving the performance of Federated Graph Learning (FGL). However, due to the limited computing and storage resources of UAVs, they may not well handle the redundant data and complex models, causing the inference inefficiency of FGL in UAV-assisted MEC systems. To address this critical challenge, we propose a novel LightWeight FGL framework, named LW-FGL, to accelerate the inference speed of classification models in UAV-assisted MEC systems. Specifically, we first design an adaptive Information Bottleneck (IB) principle, which enables UAVs to obtain well-compressed worthy subgraphs by filtering out the information that is irrelevant to downstream classification tasks. Next, we develop improved tiny Graph Neural Networks (GNNs), which are used as the inference models on UAVs, thus reducing the computational complexity and redundancy. Using real-world graph datasets, extensive experiments are conducted to validate the effectiveness of the proposed LW-FGL. The results show that the LW-FGL achieves higher classification accuracy and faster inference speed than state-of-the-art methods.

The emerging Space-Air-Ground Integrated Networks (SAGIN) empower Mobile Edge Computing (MEC) with wider communication coverage and more flexible network access. However, the fluctuating user traffic and constrained computing architecture seriously hinder the Quality-of-Service (QoS) and resource utilization in SAGIN. Existing solutions generally depend on prior knowledge or adopt static resource provisioning, lacking adaptability and resulting in serious system overheads. To address these important challenges, we propose THOAS, a novel Traffic-aware lightweight Hierarchical Offloading framework towards Adaptive Slicing-enabled SAGIN. First, we innovatively separate SAGIN into Communication Access Platforms (CAPs) and Computation Offloading Platforms (COPs). Next, we design a new self-attention-based prediction method to accurately capture the traffic changes on each platform, enabling adaptive slice resource adjustments. Finally, we develop an improved deep reinforcement learning method based on proximal clipping with dynamic confidence intervals to reach optimal offloading. Notably, we employ knowledge distillation to compress offloading policies into lightweight networks, enhancing their adaptability in resource-limited SAGIN. Using real-world datasets of user traffic, extensive experiments are conducted. The results show that the THOAS can accurately predict traffic and make adaptive resource adjustments and offloading decisions, which outperforms other benchmark methods on multiple metrics under various scenarios.

Through deploying computing resources at the network edge, Mobile Edge Computing (MEC) alleviates the contradiction between the high requirements of intelligent mobile applications and the limited capacities of mobile End Devices (EDs) in smart communities. However, existing solutions of computation offloading and resource allocation commonly rely on prior knowledge or centralized decision-making, which cannot adapt to dynamic MEC environments with changeable system states and personalized user demands, resulting in degraded Quality-of-Service (QoS) and excessive system overheads. To address this important challenge, we propose a novel Personalized Federated deep Reinforcement learning based computation Offloading and resource Allocation method (PFR-OA). This innovative PFR-OA considers the personalized demands in smart communities when generating proper policies of computation offloading and resource allocation. To relieve the negative impact of local updates on global model convergence, we design a new proximal term to improve the manner of only optimizing local Q-value loss functions in classic reinforcement learning. Moreover, we develop a new partial-greedy based participant selection mechanism to reduce the complexity of federated aggregation while endowing sufficient exploration. Using real-world system settings and testbed, extensive experiments demonstrate the effectiveness of the PFR-OA. Compared to benchmark methods, the PFR-OA achieves better trade-offs between delay and energy consumption and higher task execution success rates under different scenarios.

As an effective technique to relieve the problem of resource constraints on mobile devices (MDs), the computation offloading utilizes powerful cloud and edge resources to process the computation-intensive tasks of mobile applications uploaded from MDs. In cloud-edge computing, the resources (e.g., cloud and edge servers) that can be accessed by mobile applications may change dynamically. Meanwhile, the parallel tasks in mobile applications may lead to the huge solution space of offloading decisions. Therefore, it is challenging to determine proper offloading plans in response to such high dynamics and complexity in cloud-edge environments. The existing studies often preset the priority of parallel tasks to simplify the solution space of offloading decisions, and thus the proper offloading plans cannot be found in many cases. To address this challenge, we propose a novel real-time and Dependency-aware task Offloading method with Deep Q-networks (DODQ) in cloud-edge computing. In DODQ, mobile applications are first modeled as Directed Acyclic Graphs (DAGs). Next, the Deep Q-Networks (DQN) is customized to train the decision-making model of task offloading, aiming to quickly complete the decision-making process and generate new offloading plans when the environments change, which considers the parallelism of tasks without presetting the task priority when scheduling tasks. Simulation results show that the DODQ can well adapt to different environments and efficiently make offloading decisions. Moreover, the DODQ outperforms the state-of-art methods and quickly reaches the optimal/near-optimal performance.

Load prediction is an essential technique to improve edge system performance by proactively configuring and allocating system resources. Traditional load prediction methods obtain high prediction when handling loads exhibiting cyclical trend behavior, but they are unable to capturing highly-variable loads in edge computing environments. Existing studies fit prediction models via independent time series and output single-point real-value predictions. However, in practical edge scenarios, it is more valuable to obtain application value by utilizing the probability distribution of future loads rather than directly predicting specific values. To solve these problems, we propose an Edge Load Prediction method empowered by Deep Auto-regressive Recurrent networks (ELP-DAR). The ELP-DAR uses the time-series data of edge loads to train deep auto-regressive recurrent networks, which integrate Long Short-Term Memory (LSTM) into the S2S framework to calculate the parameters of the probability distribution at the next time-point. Therefore, the ELP-DAR can efficiently extract the essential representations of edge loads and learn their complex patterns, and the probability distribution for highly-variable edge loads can be accurately predicted. Extensive simulation experiments are conducted to validate the effectiveness of the proposed ELP-DAR method based on real-world edge load datasets. The results show that the ELP-DAR achieves higher prediction accuracy than other benchmark methods with different prediction lengths.

Mobile edge computing (MEC) relieves the latency and energy consumption of mobile applications by offloading computation-intensive tasks to nearby edges. In wireless metropolitan area networks (WMANs), edges can better provide computing services via advanced communication technologies. For improving the Quality-of-Service (QoS), edges need to be collaborated rather than working alone. However, the existing solutions of multiedge collaboration solely adopt a centralized or decentralized decision-making way of load balancing, making it hard to achieve the optimal result because the local and global conditions are not jointly considered. To solve this problem, we propose a novel two-stage decision-making method of load balancing for multiedge collaboration (TDB-EC). First, the centralized decision making is executed with global information, where a deep neural networks (DNNs)-based prediction model is designed to evaluate the range of task scheduling between adjacent edges. Next, considering the global condition of load balancing, the decentralized decision making is executed with local information, where a deep$Q$-networks (DQN)-based$Q$-value prediction model of adjustment operations is developed to evaluate the load balancing plan among edges. Finally, the objective load balancing plan is obtained via feedback control. Extensive simulation experiments demonstrate the adaptability of the TDB-EC to various scenarios of multiedge load balancing, which approximates the optimal result and outperforms three classic methods.

In the era of the Internet of Things (IoT), it is a promising way to improve system energy utility and better meet users’ requirements for Quality of Service (QoS) via integrating nonorthogonal multiple access (NOMA) and mobile-edge computing (MEC) technologies. In light of this idea, we investigate device access, subchannel division, and transmission power allocation for NOMA-enabled IoT systems. To maximize the energy utility of IoT systems while satisfying the minimum demands of IoT Devices (IoTDs) on achievable uplink data rate, a joint optimization problem is formulated with the consideration of device access, subchannel division, and transmission power allocation. Due to the nonconvexity of this problem, we propose an alternating optimization algorithm aiming to find the optimal solution. The proposed algorithm first decomposes the joint optimization problem into three subproblems through the block coordinate descent (BCD), and then obtains the near-optimal solution by solving the decomposed subproblems alternately. Extensive simulations validate our analysis for the convergence of the proposed algorithm. The numerical results demonstrate that the proposed algorithm significantly outperforms the benchmark algorithms in terms of improving system energy utility.

As the workloads and service requests in cloud computing environments change constantly, cloud-based software services need to adaptively allocate resources for ensuring the Quality-of-Service (QoS) while reducing resource costs. However, it is very challenging to achieve adaptive resource allocation for cloud-based software services with complex and variable system states. Most of the existing methods only consider the current condition of workloads, and thus cannot well adapt to real-world cloud environments subject to fluctuating workloads. To address this challenge, we propose a novel Deep Reinforcement learning based resource Allocation method with workload-time Windows (DRAW) for cloud-based software services that considers both the current and future workloads in the resource allocation process. Specifically, an original Deep Q-Network (DQN) based prediction model of management operations is trained based on workload-time windows, which can be used to predict appropriate management operations under different system states. Next, a new feedback-control mechanism is designed to construct the objective resource allocation plan under the current system state through iterative execution of management operations. Extensive simulation results demonstrate that the prediction accuracy of management operations generated by the proposed DRAW method can reach 90.69%. Moreover, the DRAW can achieve the optimal/near-optimal performance and outperform other classic methods by 3$\sim$13% under different scenarios.

Abstract The combination of unmanned aerial vehicles (UAVs) and mobile edge computing (MEC) technology breaks through the limitations of traditional terrestrial communications. The effective line-of-sight channel provided by UAVs can greatly improve the communication quality between edge servers and mobile devices (MDs). To further enhance the Quality-of-Service (QoS) of MEC systems, a multi-UAV-enabled MEC system model is designed. In the proposed model, UAVs are regarded as edge servers to offer computing services for MDs, aiming to minimize the average task response time by jointly optimizing UAV deployment and computation offloading. Based on the problem definition, a two-layer joint optimization method (PSO-GA-G) is proposed. First, the outer layer utilizes a Particle Swarm Optimization algorithm combined with Genetic Algorithm operators (PSO-GA) to optimize UAV deployment. Next, the inner layer adopts a greedy algorithm to optimize computation offloading. The extensive simulation experiments verify the feasibility and effectiveness of the proposed PSO-GA-G. The results show that the PSO-GA-G can achieve a lower average task response time than the other three baselines.

With time-varying workloads and service requests, cloud-based software services necessitate adaptive resource allocation for guaranteeing Quality-of-Service (QoS) and reducing resource costs. However, due to the ever-changing system states, resource allocation for cloud-based software services faces huge challenges in dynamics and complexity. The traditional approaches mostly rely on expert knowledge or numerous iterations, which might lead to weak adaptiveness and extra costs. Moreover, existing RL-based methods target the environment with the fixed workload, and thus they are unable to effectively fit in the real-world scenarios with variable workloads. To address these important challenges, we propose a Prediction-enabled feedback Control with Reinforcement learning based resource Allocation (PCRA) method. First, a novel Q-value prediction model is designed to predict the values of management operations (by Q-values) at different system states. The model uses multiple prediction learners for making accurate Q-value prediction by integrating the Q-learning algorithm. Next, the objective resource allocation plans can be found by using a new feedback-control based decision-making algorithm. Using the RUBiS benchmark, simulation results demonstrate that the PCRA chooses the management operations of resource allocation with 93.7 percent correctness. Moreover, the PCRA achieves optimal/near-optimal performance, and it outperforms the classic ML-based and rule-based methods by 5$\sim$∼7% and 10$\sim$∼13%, respectively.

The rapid development of the Industrial Internet of Things (IIoT) enables IIoT devices to offload their computation-intensive tasks to nearby edges via wireless base stations and thus relieve their resource constraints. To better guarantee quality-of-service, it has become necessary to cooperate multiple edges instead of letting them work alone. However, the existing solutions commonly use a centralized decision-making manner and cannot effectively achieve good load balancing among massive edges that are widely distributed in IIoT environments. This results in long decision-making time and high communication costs. To address this important problem, in this article, we propose a reinforcement learning (RL)-empowered feedback control method for cooperative load balancing (RF-CLB). First, by integrating RL and machine learning (ML) algorithms, each edge independently schedules tasks and performs load balancing between adjacent edges based on the local information. Next, through feedback control and multiedge cooperation, the objective multiedge load-balancing plan for IIoT can be found. Simulation results demonstrate that the RF-CLB chooses the adjustment operations of load balancing with 96.3% correctness. Moreover, the RF-CLB achieves the near-optimal performance, which outperforms the classic ML-based and rule-based methods by 6–9% and 10–12%, respectively.

The ever-expanding scale of cloud datacenters necessitates automated resource provisioning to best meet the requirements of low latency and high energy-efficiency. However, due to the dynamic system states and various user demands, efficient resource allocation in cloud faces huge challenges. Most of the existing solutions for cloud resource allocation cannot effectively handle the dynamic cloud environments because they depend on the prior knowledge of a cloud system, which may lead to excessive energy consumption and degraded Quality-of-Service (QoS). To address this problem, we propose an adaptive and efficient cloud resource allocation scheme based on Actor-Critic Deep Reinforcement Learning (DRL). First, the actor parameterizes the policy (allocating resources) and chooses actions (scheduling jobs) based on the scores assessed by the critic (evaluating actions). Next, the resource allocation policy is updated by using gradient ascent while the variance of policy gradient is reduced with an advantage function, which improves the training efficiency of the proposed method. We conduct extensive simulation experiments using real-world data from Google cloud datacenters. The results show that our method can obtain the superior QoS in terms of latency and job dismissing rate with enhanced energy-efficiency, compared to two advanced DRL-based and five classic cloud resource allocation methods.

Deep Neural Networks (DNNs) have become an essential and important supporting technology for smart Internet-of-Things (IoT) systems. Due to the high computational costs of large-scale DNNs, it might be infeasible to directly deploy them in energy-constrained IoT devices. Through offloading computation-intensive tasks to the cloud or edges, the computation offloading technology offers a feasible solution to execute DNNs. However, energy-efficient offloading for DNN based smart IoT systems with deadline constraints in the cloud-edge environments is still an open challenge. To address this challenge, we first design a new system energy consumption model, which takes into account the runtime, switching, and computing energy consumption of all participating servers (from both the cloud and edge) and IoT devices. Next, a novel energy-efficient offloading strategy based on a Self-adaptive Particle Swarm Optimization algorithm using the Genetic Algorithm operators (SPSO-GA) is proposed. This new strategy can efficiently make offloading decisions for DNN layers with layer partition operations, which can lessen the encoding dimension and improve the execution time of SPSO-GA. Simulation results demonstrate that the proposed strategy can significantly reduce energy consumption compared to other classic methods.

Summary Mobile edge computing (MEC) necessitates cost‐effective deployment for executing scientific workflows with different tasks and datasets, which provides computing, storage and network control at the network edge. However, the execution of scientific workflows in MEC results in heavy costs of data placement including data transmission and data storage. Although there are solutions for data placement in traditional cloud computing, they cannot effectively respond to the latency‐sensitive property of scientific workflows, which leads to the excessive costs of data placement. To cope with this problem, we combine the advantages of MEC and cloud computing and propose a genetic algorithm particle swarm optimization (GAPSO) based method to explore the optimal strategy of data placement for scientific workflows in MEC. First, a unified model of data placement is designed to explore a cost‐effective strategy, which considers the different characteristics between MEC and cloud computing as well as the impact of latency constraint on transmission costs. Next, the advantages of genetic algorithm (GA) and particle swarm optimization (PSO) are integrated to optimize the proposed model, which utilities the fast convergence of PSO and the crossover and mutation operations of GA. Simulations using real‐world scientific workflows show the effectiveness of the proposed method for reducing data placement costs in MEC.

The federated learning framework builds a deep learning model collaboratively by a group of connected devices via only sharing local parameter updates to the central parameter server. Nonetheless, the lack of transparency in the local data resource makes it prone to adversarial federated attacks, which have shown increasing ability to reduce learning performance. Existing research efforts either focus on the single-party attack with impractical perfect knowledge setting and limited stealthy ability or the random attack that has no control on attack effects. In this paper, we investigate a new multi-party adversarial attack with the imperfect knowledge of the target system. Controlled by an adversary, a number of compromised devices collaboratively launch targeted model poisoning attacks, intending to misclassify the targeted samples while maintaining stealthy under different detection strategies. Specifically, the compromised devices jointly minimize the loss function of model training in different scenarios. To overcome the update scaling problem, we develop a new boosting strategy by introducing two stealthy metrics. Via experimental results, we show that under both perfect knowledge and limited knowledge settings, the multi-party attack is capable of successfully evading detection strategies while guaranteeing the convergence. We also demonstrate that the learned model achieves the high accuracy on the targeted samples, which confirms the significant impact of the multi-party attack on federated learning systems.

Resource provisioning for cloud computing necessitates the adaptive and accurate prediction of cloud workloads. However, the existing methods cannot effectively predict the high-dimensional and highly-variable cloud workloads. This results in resource wasting and inability to satisfy service level agreements (SLAs). Since recurrent neural network (RNN) is naturally suitable for sequential data analysis, it has been recently used to tackle the problem of workload prediction. However, RNN often performs poorly on learning long-term memory dependencies, and thus cannot make the accurate prediction of workloads. To address these important challenges, we propose a deep Learning based Prediction Algorithm for cloud Workloads (L-PAW). First, a top-sparse auto-encoder (TSA) is designed to effectively extract the essential representations of workloads from the original high-dimensional workload data. Next, we integrate TSA and gated recurrent unit (GRU) block into RNN to achieve the adaptive and accurate prediction for highly-variable workloads. Using real-world workload traces from Google and Alibaba cloud data centers and the DUX-based cluster, extensive experiments are conducted to demonstrate the effectiveness and adaptability of the L-PAW for different types of workloads with various prediction lengths. Moreover, the performance results show that the L-PAW achieves superior prediction accuracy compared to the classic RNN-based and other workload prediction methods for high-dimensional and highly-variable real-world cloud workloads.

Due to the ever-changing system states and various user demands, resource allocation in cloud data center is faced with great challenges in dynamics and complexity. Although there are solutions that focus on addressing this problem, they cannot effectively respond to the dynamic changes of system states and user demands since they depend on the prior knowledge of the system. Therefore, it is still an open challenge to realize automatic and adaptive resource allocation in order to satisfy diverse system requirements in cloud data center. To cope with this challenge, we propose an advantage actor-critic based reinforcement learning (RL) framework for resource allocation in cloud data center. First, the actor parameterizes the policy (allocating resources) and chooses continuous actions (scheduling jobs) based on the scores (evaluating actions) from the critic. Next, the policy is updated by gradient ascent and the variance of policy gradient can be significantly reduced with the advantage function. Simulations using Google cluster-usage traces show the effectiveness of the proposed method in cloud resource allocation. Moreover, the proposed method outperforms classic resource allocation algorithms in terms of job latency and achieves faster convergence speed than the traditional policy gradient method.

Green communications are playing critical roles in vehicular ad hoc networks (VANETs), while the deployment of a power efficient VANET is quite challenging in practice. To add more greens into such kind of complicated and time-varying mobile network, we specifically investigate the throughput and transmission delay performances for real-time and delay sensitive services through a repeated game theoretic solution. This paper has employed Nash Equilibrium in the noncooperative game model and analyzes its efficiency. Simulation results have shown an obvious improvement on power efficiency through such efforts.

In the study of social networks, analyzing node influence and identifying influential nodes in social networks is of great theoretical and practical significance. To effectively evaluate node influence, a novel concept of influence label is introduced, which can measure node influence through two label attributes called influence level and node degree. Then, a novel influence model called LDM (Level and Degree Model) is proposed. LDM updates the influence label of each node iteratively by the quality of neighbors and the number of neighbors. A gain function that conforms to power-law distribution is also introduced to further optimize the accuracy of LDM. In the experiment part, to prove the effectiveness of LDM, LDM is compared with other typical methods by employing IC model to simulate the information diffusion process on four real-world social networks. Experimental results show that LDM can rank node influence more accurately than other methods.

Social network has become a very popular way for internet users to communicate and interact online. Users spend plenty of time on famous social networks (e.g., Facebook, Twitter, Sina Weibo, etc.), reading news, discussing events and posting messages. Unfortunately, this popularity also attracts a significant amount of spammers who continuously expose malicious behavior (e.g., post messages containing commercial URLs, following a larger amount of users, etc.), leading to great misunderstanding and inconvenience on users׳ social activities. In this paper, a supervised machine learning based solution is proposed for an effective spammer detection. The main procedure of the work is: first, collect a dataset from Sina Weibo including 30,116 users and more than 16 million messages. Then, construct a labeled dataset of users and manually classify users into spammers and non-spammers. Afterwards, extract a set of feature from message content and users׳ social behavior, and apply into SVM (Support Vector Machines) based spammer detection algorithm. The experiment shows that the proposed solution is capable to provide excellent performance with true positive rate of spammers and non-spammers reaching 99.1% and 99.9% respectively.