To improve the quality of service (QoS) for mobile users (MUs) and the quality of experience (QoE) of mobile devices (MDs), mobile edge computing (MEC) is a promising approach that offloads a part of the computing task from MDs to nearby MUs. In this paper, we study computation offloading involving multiple users and multiple base stations (BSs), where the MD that is connected to the MU is wirelessly charged and BSs are available to be selected for computation offloading. We model the process of solving an optimal computation offloading policy into a Markov decision process (MDP), in which our goal is to maximize the long-term utility performance. Therefore, a computation offloading policy is obtained based on the energy queue state, the task queue state, and the channel states between the MUs and BSs. To address the problem of high dimensionality in the state space, we decompose the MDP into a series of single-agent MDPs with reduced state spaces and apply an online local learning algorithm to learn the optimal state value functions. Inspired by the structure of the utility function, we propose an algorithm based on combining Q-function reconstruction with the post-decision state. It is proved that the proposed algorithm can converge to an optimal computation offloading policy. The experimental results show that our algorithm achieves significant performance in computation offloading and schedule compared with the other three basic policies.

Mobile edge computing meets stringent latency requirements by offloading computational tasks to edge servers. However, in dynamic and uncertain environments, assigning tasks to multiple users becomes complex. To address this challenge, we design a multi-user task offloading framework that allows users to initiate service requests in a distributed manner. Specifically, we propose an online learning offloading algorithm based on a distributed auction multi-armed bandit, which can adapt to stochastically changing environments and gradually reduce computational latency. We then transform the dynamic task allocation problem into an online multi-user multi-armed bandit problem and develop an offloading algorithm based on heterogeneous distributed multi-armed bandit (HD-MAB) to optimize user rewards subject to network latency. We demonstrate that the HD-MAB algorithm can achieve optimal task allocation, thereby providing near-optimal service performance with linear regret. Simulation results show that our offloading method performs well in optimizing latency-sensitive tasks, and that user participation in the decision-making process of the HD-MAB algorithm does not affect the asymptotic optimality of the algorithm.

Transformer-based image registration methods have achieved notable success, but they still face challenges, such as difficulties in representing both global and local features, the inability of standard convolution operations to focus on key regions, and inefficiencies in restoring global context using the decoder. To address these issues, we extended the Swin-UNet architecture and incorporated dynamic snake convolution (DSConv) into the model, expanding it into three dimensions. This improvement enables the model to better capture spatial information at different scales, enhancing its adaptability to complex anatomical structures and their intricate components. Additionally, multi-scale dense skip connections were introduced to mitigate the spatial information loss caused by downsampling, enhancing the model’s ability to capture both global and local features. We also introduced a novel optimization-based weakly supervised strategy, which iteratively refines the deformation field generated during registration, enabling the model to produce more accurate registered images. Building on these innovations, we proposed OSS DSC-STUNet+ (Swin-UNet+ with 3D dynamic snake convolution). Experimental results on the IXI, OASIS, and LPBA40 brain MRI datasets demonstrated up to a 16.3% improvement in Dice coefficient compared to five classical methods. The model exhibits outstanding performance in terms of registration accuracy, efficiency, and feature preservation.