Large Language Models (LLMs), with advanced content creation and inference capabilities, can provide immersive intelligent services to users in mobile edge networks. However, the increasing demand for real-time artificial intelligence (AI) applications aggravates the limitations of cloud-based LLMs due to the long response time. Meanwhile, Small Language Models (SLMs), which are cost-effective and locally deployable for terminal devices, can serve as an efficient supplement to LLMs for performing latency-sensitive tasks with lower generalization capability. Due to the resource constraints of edge networks and the diverse requirements of user tasks, it is critical to design an inference framework that effectively coordinates the deployment and collaboration of LLMs and SLMs. In this paper, we propose an LLM-SLM collaborative inference (LSCI) scheme under a mobile edge computing (MEC) architecture, which jointly decides where to cache models and how to offload inference tasks to balance latency, accuracy, and resource costs. To optimize inference performance subject to resource constraints, we jointly solve the inference task offloading and model caching problem in LSCI scheme. Specifically, we employ deep reinforcement learning (DRL) to select highly popular SLMs to be cached on the edge server, and distributed belief propagation technique to solve the associated inference task offloading issue. Numerical results show that the proposed LSCI scheme can achieve significant performance gain in terms of inference performance when compared with a number of baseline solutions.

Integrated Sensing and Communication (ISAC) is driving the evolution of edge intelligence. In ISAC-enabled wireless edge networks, federated learning (FL) is crucial for realizing edge intelligence by supporting the networks with privacy protection, efficient data management, and dynamic adaptability. Specifically, FL allows distributed computing nodes (e.g., sensor devices) to first train local models by using data collected or sensed via ISAC and subsequently send them to one or multiple aggregation nodes for global model collaboration. However, traditional FL frameworks face significant challenges in the ISAC scenarios. For example, the privacy sensitivity of heterogeneous sensor data and the lack of transparency in model parameter exchange make it difficult to ensure the credibility of local and global models. Sharding distributed ledger technology (DLT), which divides the ledger into smaller and manageable shards, offers a potential solution to address these challenges by utilizing multi-node trust capabilities to facilitate distributed consensus during FL training. In this paper, we propose a trusted FL framework that incorporates sharding DLT within ISAC-enabled wireless edge networks to enhance both model training and consensus performance. Specifically, we develop a theoretical model to examine the interactions between model training performance and network capacities of sensing nodes (e.g., storage, computing, and communication capabilities) based on ISAC’s real-time channel state information. Based on this theoretical model, we design a trusted clustering scheme for aggregating local models. Numerical results demonstrate that in ISAC-enabled wireless edge networks, our proposed scheme significantly increases network throughput for model transmission while ensuring optimal model learning performance compared to some classical baselines.

By leveraging the parameter-efficient fine-tuning (PEFT) technology, large pre-trained models (LPMs) can effectively capture task-specific PEFT knowledge and generalize to downstream artificial intelligence (AI) tasks emerging in wireless networks. However, the inherent system heterogeneity of wireless networks and data across users brings difficulties in combining the PEFT knowledge of users. This in turn hinders the enhancement of LPM's global generalization capability. In this paper, we propose a distributed fine-tuning (DFT) scheme based on PEFT knowledge sharing, to capture the global PEFT knowledge of downstream users by aggregating heterogeneous PEFT models. In DFT, we use low-rank adaption (LoRA) model to capture the PEFT knowledge of users, where the LoRA models can be heterogeneous for different users with specific resource constraints. Then, we further design an aggregation and iterative training strategy for the heterogeneous LoRA models. The numerical results verify the effectiveness and advantages of the proposed DFT scheme in enhancing the global generalization capability of LPM in resource-heterogeneous wireless networks.

Federated learning (FL) is integral to advancing edge intelligence by enabling collaborative machine learning. In FL-empowered edge networks, computing nodes first train local models and then send them to an or multiple aggregation node(s) for global model collaboration. However, the trustworthiness of both local and global models in conventional FL frameworks is compromised due to inadequate model security and transparency. Distributed ledger technique (DLT) can address this issue by leveraging multi-nodes trust capabilities to support distributed consensus. However, model training and consensus performance of DLT may significantly degrade due to instability and resource constraints of edge networks. Sharding technique provides an effective approach by dividing the ledger into smaller and manageable shards. In this paper, to improve model training and consensus performance, we propose a trusted FL framework by incorporating sharding DLT into FL frameworks. We construct a theoretical model to investigate the relationship between model training performance, consensus efficiency, and capacity of edge nodes regarding storage, computing and communications. Based on the theoretical model, we propose a trusted clustering scheme to aggregate local models. Numerical results show that our proposed scheme significantly improves network throughput for transmitting models while guaranteeing model learning performance in comparison with some classical baselines.

Federated Learning (FL) has been widely recognized as a promising promoter for future intelligent wireless networks, by collaboratively training a global machine learning (ML) model in a privacy-preserving manner. However, the transmission of large-scale models between clients and servers is susceptible to the limitation of available communication resources which deteriorates FL performance. Some recently proposed model quantization techniques can effectively reduce communication costs by compressing the amount of model data to be transmitted. Unfortunately, conventional quantization methods face difficulties in wireless with rapidly changing channels. In this paper, we propose a federated learning scheme with incremental model quantization and uploading mechanism, called Fed_IQ. Specifically, individual clients quantize the gradients of the local model to obtain base gradients as well as incremental gradients and send them to the server. Then the server combines the base and incremental gradients to obtain a more accurate global model where the quantization levels of gradients can be adaptively adjusted according to the instantaneous channel states. Experimental results show our proposed Fed_IQ can significantly reduce transmission delay and improve model accuracy in a wireless network compared with state-of-the-art algorithms.

Unmanned aerial vehicle (UAV) has been recognized as a key supplement for terrestrial networks to meet the stringent requirements of the forthcoming 6G networks. However, a significant challenge lies in the provisioning of differentiated services through a common UAV network without deploying individual networks for each service type. In this paper, we consider the problem of joint network slicing and UAV placement under dynamic wireless environment as well as the uncertain traffic demands. To overcome the difficulties brought by the network dynamics, we propose an intelligent hierarchical UAV slicing framework that operates at two different time-scales. At the large time-scale, we formulate the problem of inter-slice resource slicing and UAV placement as a mixed integer nonlinear program, which is solved by a decomposition algorithm. At the small time-scale, the problem of intra-slice resource adjustment is modeled as a stochastic game and a distributed learning algorithm is proposed to find the expected Nash Equilibrium. Simulation results demonstrate that the proposed framework is lightweight and outperforms a number of known benchmark algorithms in terms of throughput and transmission delay.

Federated learning (FL) has been vigorously promoted in wireless edge networks as it fosters collaborative training of machine learning (ML) models while preserving individual user privacy and data security. In conventional FL, user equipments (UEs) and an aggregator can collaboratively train a globally shared ML model by transmitting ML models instead of raw data. In wireless edge networks, the heterogeneity of multidimensional resources (e.g., computing and communication re-sources) used to transmit ML models may introduce stragglers in FL, characterized by a slow update and/or transmission of local models. The stragglers in FL can significantly degrade learning efficiency and accuracy, as the slowest UE participating in the FL can dramatically slow down entire convergence. In this paper, to alleviate the negative impact of stragglers, we propose a dynamic straggler-aware clustering based FL mechanism, called FeDSC, via adaptive UE clustering. Specifically, we first group participating UEs into multiple clusters based on their computing capability and available wireless resources. Then, we propose an adaptive UE selection scheme to synchronously update the cluster aggregation model. Meanwhile, an edge server performs global aggregation of different cluster models in an asynchronous time-triggered manner. Numerical results show that our proposed FeDSC mechanism can achieve significant performance improvement in terms of training time and model accuracy in comparison to classical FL benchmarks.

Unmanned aerial vehicle (UAV) has been recognized as a key supplement for terrestrial networks to meet the stringent requirements of the forthcoming 6G networks. However, a significant challenge lies in providing differentiated services through a common UAV network, without the need to deploy individual networks for each service type. In this paper, we consider the problem of joint network slicing and UAV deployment under dynamic wireless environments as well as the uncertain traffic demands. To overcome the challenges posed by the network dynamics, we propose an intelligent hierarchical UAV slicing framework that operates at two different time-scales. At the large time-scale, the problem of inter-slice resource slicing and UAV deployment is formulated as a mixed integer nonlinear program, and a decomposition technique is applied to resolve it. At the small time-scale, the problem of intra-slice resource adjustment is modeled as a stochastic game and a distributed learning algorithm is proposed to find its Nash Equilibrium. Simulation results demonstrate that the proposed framework is lightweight and outperforms a number of known benchmark algorithms in terms of system utility, throughput and transmission delay.

Federated learning (FL) has been vigorously promoted in wireless edge networks as it fosters collaborative training of machine learning (ML) models while preserving individual user privacy and data security. In conventional FL, user equipment (UE) and an aggregator can collaboratively train shared global ML models by transmitting interactive ML models. In wireless edge networks, heterogeneity of multi-dimensional resources (e.g., computing and communication resources) used to train and transmit FL models may introduce stragglers, characterized by a slow update and/or transmission of local models. The stragglers can significantly degrade learning performance of FL, as the slowest participating UE can dramatically slow down the entire convergence. In this paper, to alleviate the negative impact of stragglers, we propose a dynamic straggler-aware clustering based FL (FeDSC) mechanism via adaptive UE clustering. Specifically, we first group participating UEs into multiple clusters based on their available computing and wireless resources. Then, we propose an adaptive clustering scheme to synchronously update the cluster aggregation model. Meanwhile, an edge server performs global aggregation of different cluster models in an asynchronous manner. Finally, we theoretically demonstrate the convergence of our proposed mechanism via numerical results. Numerical results show that our proposed mechanism can effectively reduce training time and wireless bandwidth consumption, while improving training efficiency and guaranteeing learning accuracy.

To accommodate the growing number of heterogeneous radios with limited wireless resources, symbiotic communication (SC) inspired by biology has been recently proposed to establish a symbiotic radio (SR) ecosystem. In this SR ecosystem, through collaboratively optimizing service/resource exchange policies, radios can coevolve like organisms, thus enabling various radio resources (such as spectrum, energy, and computing power) to complement each other. However, one critical challenge is securing a trusted coevolution environment in an SR ecosystem since the SRs with different network operators should coevolve under unreliable wireless links with complex electromagnetic interference. Moreover, multidimensional resources participated and a wide array of service requirements pose additional challenges to service/resource exchange decision-making across massive SRs. In this paper, we propose a Blockchain-empowered Intelligent cOevolution scheme for SRs, named BIO-SR. Specifically, BIO-SR exploits the digital acyclic graph (DAG) blockchain consensus in securing a trusted environment of SRs and applies deep reinforcement learning (DRL) in service exchange decision-making. The simulation results show that the BIO-SR scheme outperforms conventional solutions in terms of transmission rate and latency under both non-attack and malicious attack scenarios.

High frequency directional communication is considered as a key technology to improve the performance of mobile Ad Hoc networks owing to its advantages in terms of communication distance and interference. Neighbor discovery plays a key role for efficient routing and topology control in mobile Ad Hoc networks. It is also a very challenging issue due to the use of directional antennas and movement of mobile nodes. Beam configuration is the key step of neighbor discovery in mobile Ad Hoc networks with directional antennas. Therefore, it is imperative to develop an efficient beam configuration algorithm to reduce the neighbor discovery latency. In this paper, we propose a novel beam configuration algorithm based on personalized federated learning. Considering the characteristics of mobile Ad Hoc networks (e.g., dynamic topology and directional communication), we use Deep Deterministic Policy Gradient (DDPG) as the local model of federated learning. Since the local data of ad hoc network nodes is heterogeneous, Model Agnostic Meta Learning (MAML) is applied to personalize the federated learning. Numerical results demonstrate that our proposed algorithm has better performance than some baseline algorithms.

Unmanned aerial vehicles (UAVs) are capable of serving as aerial base stations (BSs) for providing dynamic coverage and connectivity extension for the sixth-generation (6G) wireless networks. While flexibility is provided, the deployment of the UAV swarms and the associated resource allocation bring challenging issues due to dynamic nature of UAVs and difficulty in obtaining global user information. In this paper, we propose an adaptive and flexible joint UAV deployment and resource allocation scheme by exploiting a personalized federated deep reinforcement learning framework, called PFRL, with aim to maximize the long-term network throughput while enforcing user privacy and adapting to time-varying network states. To allow UAVs to make real-time decisions on resource allocation and position adjustment based on local observations while achieving a global optimal solution, we incorporate deep reinforcement learning (DRL) into federated learning framework. Specifically, we use DRL to train a local model and a personalized model on UAVs, and employ a two-level parameter aggregation scheme on a leading UAV to form a global model. The personalized model can adapt to specific environments, while exploiting the generalization of global model to accelerate the learning convergence. Numerical results show that the proposed PFRL scheme can achieve significant performance gain in terms of network throughput and convergence in comparison with some state-of-art solutions.

Mobile crowd sensing (MCS) is a promising paradigm which leverages sensor-embedded mobile devices to collect and share data. The key challenging issues in designing an MCS system include selecting appropriate users to participate in a specific sensing task and designing efficient data sensing and transmission policies for data aggregation. In mobile edge networks, the limitation on network resources including bandwidth and energy affects the design of MCS significantly. Specifically, the limited resources affect whether and how to select users for a sensing task, and the bandwidth allocated to a user affects its data sensing and transmission policies. Since user selection, bandwidth allocation, data sensing and transmission are closely coupled issues in MCS, we focus on designing a unified framework for joint sensing and communication in this paper, by jointly optimizing the aforementioned four policies under resource constraints. Simulation results show that the proposed unified framework significantly outperforms several baseline solutions without considering wireless link vulnerability and/or resource limitations.

Mobile crowdsensing (MCS) is a promising paradigm where sensor-embedded mobile devices are exploited for collecting and sharing environmental data. In MCS, the participating mobile devices sense the environment, collect the data, (pre-)process the data and transmit the data or pre-processing results to the server for further processing. In wireless edge networks, transmission and/or processing of sensed data may be unsuccessful due to the unstable wireless channels, limited bandwidth, energy and computation resources. To optimize the MCS performance, it is imperative to jointly consider the data sensing, processing and transmission for MCS system design. In this paper, we propose a joint sensing, communication and computation (JSCC) framework for multi-dimensional resource constrained MCS systems. We formulate the JSCC design as an optimization problem, by jointly controlling the data sensing, transmission and computation offloading schemes in the system. Simulation results show that the proposed JSCC framework significantly outperforms several baseline solutions without jointly considering data sensing-transmission-computation and/or multi-dimensional resource limitations.

Federated learning (FL) has become a promising technique for developing intelligent wireless networks. In traditional FL paradigms, local models are usually required to be homogeneous for aggregation. However, due to heterogeneous models coming with wireless sysTem heterogeneity, it is preferable for user equipments (UEs) to undertake appropriate amount of computing and/or data transmission work based on sysTem constraints. Meanwhile, considerable communication costs are incurred by model training, when a large number of UEs participate in FL and/or the transmitted models are large. Therefore, resource-efficient training schemes for heterogeneous models are essential for enabling FL-based intelligent wireless networks. In this paper, we propose an adaptive quantization scheme based on ensemble distillation (AQeD), to facilitate heterogeneous model training. We first partition and group the participating UEs into clusters, where the local models in specific clusters are homogeneous with different quantization levels. Then we propose an augmented loss function by jointly considering ensemble distillation loss, quantization levels and wireless resources constraints. In AQeD, model aggregations are performed at two levels: model aggregation for individual clusters and distillation loss aggregation for cluster ensembles. Numerical results show that the AQeD scheme can significantly reduce communication costs and training time in comparison with some state-of-the-art solutions.

Mobile crowd sensing (MCS) is a promising paradigm which leverages sensor-embedded mobile devices to collect and share data. To perform a sensing task in MCS, appropriate participating users are selected first, and efficient data sensing and transmission policies are then designed for data aggregation. In mobile edge networks, network resource availability affects how to select the participating users, and the bandwidth allocated to a user affects its process of data sensing and transmission. Since user selection, bandwidth allocation, data sensing and transmission are closely coupled issues in a resource constrained MCS system, we focus on designing a joint sensing and communication framework in this paper, by jointly optimizing the aforementioned four policies under resource constraints. Specifically, the optimal data sensing and transmission policies are first derived under a given user selection and bandwidth allocation scheme. Then the user selection and bandwidth allocation are optimized based on dynamic programming. Simulation results show that the proposed mechanism significantly outperforms several baseline solutions without considering wireless link vulnerability and/or resource limitations.

Mobile crowdsensing (MCS) is a promising paradigm where sensor-embedded mobile devices are exploited for collecting and sharing environmental data. In MCS, the participating mobile devices collect the required data, (pre-)process the data and transmit the data and/or pre-processing results to the server for further processing. In wireless edge networks, transmission and/or processing of sensed data may be unsuccessful due to unstable wireless channels, limited bandwidth, energy and computation resources. To optimize MCS performance, it is imperative to jointly design the data sensing, processing and transmission policies under resource constraints. In this paper, we propose a joint sensing, communication and computation (JSCC) framework for multi-dimensional resource constrained MCS systems. We formulate the JSCC design as an optimization problem, by jointly controlling the data sensing, transmission and computation offloading processes in the system. Simulation results show that the proposed JSCC framework significantly outperforms several baseline solutions without jointly considering data sensing-transmission-computation and/or multi-dimensional resource limitations.

Federated learning (FL) has recently become one of the most acknowledged technologies in promoting the development of intelligent edge networks with the ever-increasing computing capability of user equipment (UE). In traditional FL paradigm, local models are usually required to be homogeneous for aggregation to achieve an accurate global model. Moreover, considerable communication cost and training time may be incurred in resource-constrained edge networks due to a large number of UEs participating in model transmission and the large size of transmitted models. Therefore, it is imperative to develop effective training schemes for heterogeneous FL models, while reducing communication cost as well as training time. In this paper, we propose an adaptive quantization scheme based on ensemble distillation (AQeD) for FL to facilitate personalized quantized model training over heterogeneous local models with different size, structure, and quantization level, etc. Specifically, we design an augmented loss function by jointly considering distillation loss function, quantization values and available wireless resources, where UEs train their local personalized machine learning models and send the quantized models to a server. Based on local quantized models, the server first performs global aggregation for cluster ensembles and then sends the aggregated model of the cluster back to the participating UEs. Numerical results show that our proposed AQeD scheme can significantly reduce communication cost as well as training time in comparison with some known state-of-the-art solutions.

Cache-enabled device-to-device (D2D) communication is a potential approach to tackle the resource shortage problem. However, public concerns of data privacy and system security still remain, which thus arises an urgent need for a reliable caching scheme. Fortunately, federated learning (FL) with a distributed paradigm provides an effective way to privacy issue by training a high-quality global model without any raw data exchanges. Besides the privacy issue, blockchain can be further introduced into the FL framework to resist the malicious attacks occurred in D2D caching networks. In this study, we propose a double-layer blockchain-based deep reinforcement FL (BDRFL) scheme to ensure privacy-preserved and caching-efficient D2D networks. In BDRFL, a double-layer blockchain is utilized to further enhance data security. Simulation results first verify the convergence of the BDRFL-based algorithm, and then demonstrate that the download latency of the BDRFL-based caching scheme can be significantly reduced under different types of attacks when compared to some existing caching policies.

Federated learning (FL) has recently become one of the hottest focuses in wireless edge networks with the ever-increasing computing capability of user equipment (UE). In FL, UEs train local machine learning models and transmit them to an aggregator, where a global model is formed and then sent back to UEs. In wireless networks, local training and model transmission can be unsuccessful due to constrained computing resources, wireless channel impairments, bandwidth limitations, etc., which degrades FL performance in model accuracy and/or training time. Moreover, we need to quantify the benefits and cost of deploying edge intelligence, as model training and transmission consume certain amount of resources. Therefore, it is imperative to deeply understand the relationship between FL performance and multiple-dimensional resources. In this paper, we construct an analytical model to investigate the relationship between the FL model accuracy and consumed resources in FL empowered wireless edge networks. Based on the analytical model, we explicitly quantify the model accuracy, available computing resources and communication resources. Numerical results validate the effectiveness of our theoretical modeling and analysis, and demonstrate the trade-off between the communication and computing resources for achieving a certain model accuracy.

Cache-enabled device-to-device (D2D) communication has been widely deemed as a promising approach to tackle the unprecedented growth of wireless traffic demands. Recently, tremendous efforts have been put into designing an efficient caching policy to provide users better quality of service. However, public concerns of data privacy still remain in D2D cache sharing networks, which thus arises an urgent need for a privacy-preserved caching scheme. In this study, we propose a double-layer blockchain-based federated learning (DBFL) scheme with the aim of minimizing the download latency for all users in a privacy-preserving manner. Specifically, in the sublayer, the devices within the same coverage area run a federated learning (FL) to train the caching scheme model for each area separately without exchange of local data. The model parameters for each area are recorded in sublayer chains with Raft consensus mechanism. Meanwhile, in the main layer, a mainchain based on practical Byzantine fault tolerance (PBFT) mechanism is used to resist faults and attacks, thus securing the reliability of FL updates. Only the reliable area models authorized by the mainchain are utilized to update the global model in the main layer. Numerical results show the convergence, as well as the gain of download latency of the proposed DBFL caching scheme when compared with several traditional schemes.

Network Slicing (NS) has been widely identified as a key architectural technology for 5G-and-beyond systems by supporting divergent requirements sustainably. With the widespread of emerging smart devices, access control becomes an essential yet challenging issue in NS-based wireless networks due to the device-base station (BS)-NS three-layer association relationship. Meanwhile, stringent data security and device privacy concerns are increasing dramatically. In this paper, we propose an efficient access control scheme for radio access network (RAN) slicing by exploiting a federated deep reinforcement learning framework, called FDRL-AC, to improve network throughput and communication efficiency while enforcing the data security and device privacy. Specifically, we use deep reinforcement learning to train local model on devices, where horizontally federated learning (FL) is employed for parameter aggregation on BS, while vertically FL is employed for feature aggregation on the encrypted party. Numerical results show that the proposed FDRL-AC scheme can achieve significant performance gain in terms of network throughput and communication efficiency in comparison with some state-of-art solutions.

Network Function Virtualization has been widely acknowledged as one of the fundamental technologies for 5G and beyond by consolidating network functions into general-purpose hardware. To support a specific type of service, a virtualized network topology, such as Service Function Chain, is constructed by logically connecting a set of virtual network functions (VNFs). Meanwhile, Mobile Edge Computing (MEC) provides cloud resources at the edge of networks, meeting the stringent service requirements of many emerging mobile applications. With the widespread of new compute-intensive and Internet of Things applications, the amount of service flows in edge networks is rapidly increasing, causing network congestion easily because of the sinking of the computing capabilities. Moreover, due to the limited coverage of edge servers and erratic user mobility, it is difficult to maintain satisfactory service performance. Therefore, in order to support diverse services with various Quality of Service requirements, an online dynamic VNF migration is imperative in mobile networks. In this paper, we investigate the NF migration in softwarization based mobile networks with MEC, with the aim to minimize the number of link congestion in the networks. We first formulate the link congestion problem as a multidimensional knapsack problem, which is proved NP-hard. Then we resort to deep reinforcement learning to solve the online migration problem. Numerical results show that the proposed migration strategy can significantly reduce the number of link congestion and end-to-end service delay in comparison with the state-of-the-art solutions.

Software Defined Networking (SDN) and Network Function Virtualization (NFV) have been widely acknowledged as the fundamental architectural technologies for 5G and beyond networks by consolidating network functions into general-purpose hardware. Meanwhile, emerging Mobile Edge Computing (MEC) technology provides a promising solution to fulfill service requirements on high reliability and low latency, by extending cloud computing to the edge of networks. To provision a specific type of service, a virtualized network topology, such as Service Function Chain or Service Function Graph is constructed by logically connecting a set of virtual network functions (VNFs). In SDN/NFV-based mobile networks with the MEC cluster, it is imperative to develop an effective VNF deployment strategy to support various services with diverse Quality of Service requirements. In this paper, we propose a VNF deployment strategy for clustered MEC, including VNF placement and routing schemes, with the aim to minimize the average delay of service flows. We first formulate the problem as a two-dimensional knapsack problem, which is NP-Hard. To provide an efficient solution, we develop an improved genetic simulated annealing algorithm. Numerical results show that the proposed strategy can significantly reduce the end-to-end service delay in comparison with the state-of-the-art solutions.