Deep neural network (DNN) model partitioning and pruning have proven to be effective methods for enhancing resource efficiency and reducing inference delay by strategically allocating DNN workloads across heterogeneous edge and cloud infrastructures. Nevertheless, the heterogeneous nature of resources complicates the deployment of DNN in mobile edge-cloud computing (MEC) networks. In this paper, we present an innovative framework for collaborative DNN inference in MEC networks by integrating fine-grained model partitioning and magnitude-based pruning. However, the joint model partitioning and pruning policy presents significant challenges due to the inherently coupled and mutually influential nature. To address it, we adopt Long Short-Term Memory (LSTM) networks as action generation controllers to generate discrete actions for model partitioning and pruning alternately. After that, we adopt the policy gradient algorithms to optimize the LSTM-generated actions with a moving average according to the Monte Carlo estimate. By directly optimizing the policy function, the proposed framework enhances the efficiency and stability of action space exploration, yielding faster convergence and improved inference performance. Experimental results on standard datasets indicate that the proposed framework outperforms state-of-the-art approaches, achieving an 8.247% increase in system reward and an average reduction of 27.313% in total delay within the considered MEC networks.

Deploying federated learning (FL) in mobile edge computing (MEC) networks has become a prevalent approach to distributed learning. However, the inherent heterogeneity in computing, transmission and data on edge devices (EDs) brings challenges in improving training efficiency and speed. Existing approaches primarily focus on increasing batch sizes or employing adaptive batching to expedite convergence, but often overlook the generalization ability of the model. In this paper, we propose an energy-efficient adaptive batching approach for FL in MEC networks, aiming at minimizing the energy consumption by balancing training efficiency and speed. Initially, we exploit the relationship between batch size and loss improvement while determining the optimal learning rate corresponding to the batch size and understanding the correlation among loss improvement, learning rate, and gradient noise scale (GNS). Then we dynamically adjust the batch size based on the GNS and propose a low-complexity approach for measuring GNS. Finally, we fine-tune the batch size by assessing gradient similarity on each ED to ensure an optimal level of gradient noise during training, thereby enhancing the model's generalization. Experiment results demonstrate the effectiveness of our approach with an approximate 50% and 20% reduction in energy consumption and time consumption compared with existing approaches.

Federated Learning (FL) has significant potential to protect data privacy and mitigate network burden in mobile edge computing (MEC) networks. However, due to the system and data heterogeneity of mobile clients (MCs), client selection and bandwidth allocation is key for achieving cost-efficient FL in MEC networks with limited bandwidth. To address these challenges, we investigate the issue of joint client selection and bandwidth allocation for reducing the cost (i.e., latency and energy consumption) of FL training. We formulate the problem and decompose it into a holistic subproblem to reduce the number of rounds and a partial subproblem to reduce the costs of FL each round. We propose a joint class-balanced client selection and bandwidth allocation (CBCSBA) framework to address the whole problem. Specifically, for the holistic subproblem, CBCSBA combines MCs into groups, each having data distribution as close as possible to class-balanced distribution; For the partial subproblem, CBCSBA reduces costs by exploratively selecting a group and sequentially optimizing the latency and energy consumption of MCs within the group. Experimental results show that CBCSBA outperforms the baseline frameworks in reducing latency by 28.2% and energy consumption by 25.3% on average in the considered four datasets.

To address network congestion and data privacy concerns, federated learning (FL) that combines multiple clients and a parameter server has been widely used in mobile edge computing (MEC) networks to process the abundant data generated by mobile clients. However, the existing client sampling methods do not adequately consider the data heterogeneity and system heterogeneity. Parameter server selects inappropriate clients to participate in the FL training process. This inevitably leads to slower convergence of the global model and higher energy consumption. In this paper, we design a client sampling model with the goal of selecting suitable clients to improve the energy efficiency of FL in heterogeneous MEC networks. Then we propose an energy-efficient client sampling strategy by quantifying the communication capability, computation capability and data quality of clients. Based on the quantization results, clients are assigned with a corresponding sampled probability. Simulation results show that our proposed strategy can effectively accelerate the convergence of the global model and reduce the energy consumption compared with the baseline schemes.

Mobile edge-cloud computing utilizes the computing resources of edge devices and cloud servers to execute complex deep neural networks (DNNs) for collaborative inference. However, many existing collaborative inference methods do not fully consider the limited resources of edge devices, resulting in high inference latency. In this paper, we design an integrated computational framework that combines model partition and compression to reduce inference latency. Specifically, we partition a DNN model at the middle layer and deploy the previous layer on the edge device and the subsequent layer on the cloud server respectively. We propose a collaborative dual-agent reinforcement learning algorithm called CPCDRL to determine partition point and compression ratios. It enables adaptive adjustments of compression ratios based on various partition points, with the overarching goal of minimizing the inference latency across the entire DNN model. The proposed algorithm can significantly reduce computational latency while minimizing accuracy loss compared to the baseline schemes.

Model partitioning is a promising technique for improving the efficiency of distributed inference by executing partial deep neural network (DNN) models on edge servers (ESs) or Internet-of-Things (IoT) devices. However, due to heterogeneous resources of ESs and IoT devices in mobile edge computing (MEC) networks, it is non-trivial to guarantee the DNN inference speed to satisfy specific delay constraints. Meanwhile, many existing DNN models have a deep and complex architecture with numerous DNN blocks, which leads to a huge search space for fine-grained model partitioning. To address these challenges, we investigate distributed DNN inference with fine-grained model partitioning, with collaborations between ESs and IoT devices. We formulate the problem and propose a multi-task learning based asynchronous advantage actor-critic approach to find a competitive model partitioning policy that reduces DNN inference delay. Specifically, we combine the shared layers of actor-network and critic-network via soft parameter sharing, and expand the output layer into multiple branches to determine the model partitioning policy for each DNN block individually. Experiment results demonstrate that the proposed approach outperforms state-of-the-art approaches by reducing total inference delay, edge inference delay and local inference delay by an average of 4.76%, 10.04% and 8.03% in the considered MEC networks.

The development of deep learning and edge computing provides rapid detection capability for surface defects. However, components produced in actual industrial manufacturing environments often have tiny surface defects and training data for each specific defect type is limited. Meanwhile, network resources at the edge of industrial networks are difficult to guarantee. It is challenging to train a proper surface defect detection model for each specific surface defect type and provide a real-time surface defect detection service. To address the challenge, in this paper, we propose a real-time surface defect detection framework based on transfer learning with multi-access edge-cloud computing (MEC) networks. Furthermore, we improve the original YOLO-v5s framework by introducing the spatial and channel attention mechanism, and adding an additional detection head to enhance the detection ability on tiny surface defects. Evaluation results demonstrate that the proposed framework has superior performance in terms of improving detection accuracy and reducing detection delay in the considered MEC network.

To break data silos and address the challenge of green communication, federated learning (FL) is widely used at network edges to train deep learning models in mobile edge computing (MEC) networks. However, many existing FL algorithms do not fully consider the dynamic environment, resulting in slower convergence of the model and larger training energy consumption. In this paper, we design a dynamic asynchronous federated learning (DAFL) model to improve the efficiency of FL in MEC networks. Specifically, we dynamically choose a certain number of mobile devices (MDs) by their arrival order to participate in the global aggregation at each epoch. Meanwhile, we analyze the energy consumption model of local update and upload update, and formulate the problem as a dynamic sequential decision problem to minimize the energy consumption, which is NP-hard. To address it, we propose an energy-efficient algorithm based on deep reinforcement learning named DDAFL, to intelligently determine the number of MDs participating in global aggregation according to the state of MEC networks at each epoch. Compared with baseline schemes, the proposed algorithm can significantly reduce energy consumption and accelerate model convergence.

Mobile edge computing (MEC) provides a new ecosystem that enables cloud computing capabilities at the edge of mobile networks, which is characterized by ultra-low latency and high bandwidth as well as real-time access to radio network information leveraged by applications. Nevertheless, various challenges, especially the decision-making issues for task offloading, are yet to be properly addressed. In this paper, leveraging the insight from the relative evaluation method, we propose a metric to quantify the benefit on users’ service experience enhancement by task offloading. Meanwhile, by comprehensively considering the energy cost, time cost and users’ service experience enhancement throughout the task offloading process, we formulate the task offloading decision-making problem as a two-dimensional knapsack loading problem to maximize the cost efficiency of task offloading. To solve the optimization problem more efficiently, we propose a suboptimal heuristic algorithm with polynomial-time complexity. Compared with four baseline algorithms, simulation results demonstrate the cost efficiency improvement of our proposed scheme.

With the tremendous growth of mobile data traffic generated by various devices such as smartphones, smartpads and wearable devices, it is necessary for mobile network operators to introduce revolutionary networking techniques, thereby satisfying service requirements of mobile users. Recently, mobile edge computing (MEC) has been regarded as an effective technique to alleviate the traffic burden on backhaul networks. In this paper, we investigate the issue of mobility-aware content caching and user association for ultra-dense MEC networks by minimizing the system costs. The problem is formulated as a complex pure integer nonlinear programming, which is NP-hard. To address the original long-term optimization problem, we decompose it into a series of one-slot subproblems, and then optimize the short-term subproblem in two phases (i.e., content caching and user association). We further propose a mobility-aware online caching algorithm to achieve content caching, and a lazy re-association algorithm to determine user association based on matching theory. Trace-driven evaluation results demonstrate that the proposed framework has superior performance on reducing system costs.

Recently, mobile edge computing has received widespread attention, which provides computing infrastructure via pushing cloud computing, network control, and storage to the network edges. To improve the resource utilization and Quality of Service, we investigate the issue of task offloading for End-EdgeCloud orchestrated computing in mobile networks. Particularly, we jointly optimize the server selection and resource allocation to minimize the weighted sum of the average cost. A cost minimization problem is formulated underjoint the constraints of cache resource and communication/computation resource of edge servers. The resultant problem is a Mixed-Integer Non-linear Programming, which is NP-hard. To tackle this problem, we decompose it into simpler subproblems for server selection and resource allocation, respectively. We propose a low-complexity hierarchical heuristic approach to achieve server selection, and a Cauchy-Schwards Inequality based closed-form approach to efficiently determine resource allocation. Finally, simulation results demonstrate the superior performance of the proposed scheme on reducing the weighted sum of the average cost in the network.