In 6G-enabled smart manufacturing factories, software systems rapidly customize and deploy workflows through virtualization, modularization, and servitization. This enables flexible and efficient production scheduling. However, uncertainties such as equipment failures, changing task priorities, and dynamic resource demands are significant workflow execution challenges. This paper presents a method based on graph attention networks and deep reinforcement learning for workflow scheduling (GRWS), which is aimed at optimizing the workflow execution time and the associated cost, increasing the efficiency of task scheduling, and supporting flexible production manufacturing. First, topological sorting is applied to determine task dependencies, and tasks are matched with the corresponding containers to construct a container queue. By calculating the sub-deadlines of each container, the execution order of the containers is prioritized to ensure that tasks are completed efficiently within the specified time frame. Second, a reinforcement learning framework combined with a graph attention network is used to optimize aggregation and collaboration between machine nodes. This method minimizes the machine leasing cost while ensuring that the container-to-machine scheduling process meets the appropriate deadlines, thereby increasing the system’s overall efficiency. Third, to address uncertainties such as sudden workflow arrivals and machine failures, a dynamic adjustment strategy is designed to increase the robustness of the system. Finally, experiments show that compared with the existing state-of-the-art algorithms under various conditions, the proposed method reduces the incurred leasing costs by approximately 7.1%, increases the success rate by approximately 2.6%, and reduces the deadline violation rate by approximately 40.1%.

Accurate medical image segmentation is crucial in clinical applications. The existing Swin-UNet model overcomes the limitations of traditional Transformers in handling local details and high-frequency information by introducing the Swin Transformer. However, there is still room for improvement in feature fusion and multi-scale information processing. U-shaped segmentation models typically fuse features from different levels using element-wise addition or feature concatenation. However, these fusion strategies fail to fully exploit the complementary nature of features at different levels, resulting in blurry lesion boundary segmentation and thus reducing overall segmentation performance. To overcome these limitations, we propose MedSwin-UNet, a general framework for multi-modal medical image segmentation. Specifically, we equip the model with two key modules: Cross-modal Feature Injection (CFI) and Inter-level Multi-scale Perception (IMP). The CFI module effectively merges local and global features by combining CNN encoder layers with the Swin-UNet encoder, utilizing multi-modal information to enhance the model’s performance in complex medical image tasks. The core component of the IMP module is Deep Feature Enhancement (DFE), which employs a pyramid-style feature enhancement strategy to improve the fusion of deep and shallow features in Swin-UNet, thereby strengthening the model’s ability to capture both image details and overall structure. Our study surpasses most existing state-of-the-art methods on four different medical image segmentation task datasets, achieving excellent performance across multiple evaluation metrics.

Human motion prediction aims to forecast future motions based on historical motion sequences. Graph Convolutional Networks (GCNs) are widely used in this field. However, the high computational cost of GCN-based models often makes them challenging to analyze and modify. To address this, some models based on Multi-Layer Perceptrons (MLPs) have been proposed, which are simpler but entirely discard graph structure information. In this paper, we propose Spatio-Temporal Prior Graph Convolution Mixer (STPGC-Mixer), which leverages an improved space-time separable graph convolution to significantly reduce the model’s parameter size, while integrating it with the Mixer architecture to independently capture the temporal and spatial correlations in human motion sequences. We further explore and utilize implicit prior information in the human kinetic structure to guide motion prediction. By incorporating prior information, we rationally combine shared and non-shared adjacency matrices to propose a graph construction method that is more suitable for human motion prediction. Compared to other baselines, our model achieves superior results with fewer parameters on the Human3.6M, CMU-Mocap, and 3DPW datasets, particularly excelling in short-term predictions.

Video Moment Retrieval (VMR) is a key cross-modal task with broad theoretical and practical applications. While fully supervised methods deliver strong performance, they are constrained by the high cost of temporal boundary annotations. Weakly supervised methods mitigate this issue but suffer from limited accuracy due to coarse supervision. Recently, point-supervised approaches that leverage single-frame annotations as a cost-effective alternative have emerged as a promising paradigm. However, these methods often fail to leverage annotated frames for cross-modal semantic alignment. Additionally, they overlook global video structures and hierarchical segment relationships, leading to suboptimal retrieval accuracy under sparse supervision. To address these challenges, we propose the Adaptive Dual-Stage Tree Construction (ADTC) model, a novel framework designed specifically for point-supervised VMR. First, the model introduces a dual-stage hypothesis tree architecture that seamlessly integrates local and global trees, enabling the effective modeling of semantic relationships across multiple temporal scales. Second, it incorporates frame clustering and scene segmentation to extract the structural characteristics of video content, providing a foundation for comprehensive node relevance evaluation and an adaptive merging control strategy to optimize tree construction. Third, a hierarchical adaptive tree pruning strategy is implemented, combined with a novel proposal selection mechanism for distinguishing between positive and negative samples. These components are jointly optimized through a multilevel loss function, enabling enhanced semantic alignment and retrieval performance. The experimental results demonstrate that ADTC achieves state-of-the-art performance on the Charades-STA and ActivityNet Captions datasets under the point-supervised setting. On Charades-STA, it attains R@1 scores of 50.28% at IoU = 0.5 and 34.79% at IoU = 0.7, outperforming existing point-supervised methods. On ActivityNet Captions, it achieves R@1 scores of 65.02% at IoU = 0.3 and 46.13% at IoU = 0.5, establishing new benchmarks. Notably, it approaches fully supervised performance while significantly reducing annotation costs. Ablation studies further confirm the effectiveness of each model component.

IPv6 is instrumental in the ultra-large-scale intelligent computing interconnection system, yet its integration is not without challenges. The vast IPv6 address space renders traditional brute-force scanning methods infeasible, with the considerable time and resource consumption severely impacting the integration of supercomputing capabilities. This also affects the accuracy and efficiency required in high-performance computing environments. Consequently, it becomes necessary to develop new scanning technologies to address the unique challenges presented by the expansive IPv6 address space. Our novel approach, 6Diffusion-LM, transforms IPv6 scanning by fusing diffusion and linguistic models. Utilizing the Transformer architecture, it excels at extracting key features from IPv6 addresses and employs clustering algorithms to organize them effectively. Building upon the BERT pre-trained language model, 6Diffusio-LM integrate a noise mechanism that encapsulates the inherent randomness and unpredictability inherent in IPv6 address generation. The model then refines this process by progressively eliminating noise to yield precise and clear IPv6 addresses. Additionally, our proprietary embedding method enhances the generation process, ensuring higher quality addresses. Our experiments demonstrate that 6Diffusion-LM surpasses conventional methods, boasting a remarkable hit rate improvement to 43.53%

Graph Neural Networks (GNNs) have achieved significant success in graph-related tasks, particularly in scenarios involving graphs with comprehensive information. Nevertheless, the performance of GNNs is often hindered in real-world applications due to the presence of incomplete graph data, characterized by fragmented structures, missing features, and insufficient labels. Previous research in this field has predominantly concentrated on augmenting one specific individual aspect of such weak information, thereby neglecting the holistic graph learning challenge. In this paper, we introduce a new framework, named MI-GNN, standing for Multi-Interaction Graph Neural Network, which is innovatively designed to facilitate message passing across graphs while adeptly harnessing the interplay and synergies among various types of information in graph learning tasks. This approach holistically integrates graph structures, features, and labels into a cohesive learning model, thus significantly enhancing graph learning capabilities for graphs with incomplete data. Furthermore, to achieve consistency in these enhanced tasks, our framework integrates multi-task learning to address the challenges of Graph Learning with Weak Information (GLWI). We adopt an adaptive alignment strategy, dynamically assigning weights to each learning task in every iteration. Empirical evaluation of eight publicly accessible datasets demonstrates that our MI-GNN framework achieves state-of-the-art performance in handling a range of weak information scenarios in graph learning.

The widespread adoption of Industrial Internet of Things (IIoT)-based applications has driven the emergence and development of cloud-related computing paradigms with the ability to seamlessly leverage cloud resources. Heterogeneous resources, mobility factors in IoT, and dynamic behavior make it challenging for the corresponding virtual machine (VM) scheduling problem to address the processing effectiveness of application requests in these kinds of cloud environments. Based on reinforcement learning theory, this article proposes an online VM scheduling scheme (OSEC) for joint energy consumption and cost optimization that divides the scheduling process into two parts: VM allocation and VM migration. First, all the VMs and the physical machines (PMs) are regarded as a set of states and actions in the cloud environment, and the Q-learning feedback is used to achieve the iterative computation of Q-values to obtain the optimal parallel allocation sequence for multiple VMs. Then, VMs are migrated among the active PMs according to a grouping policy and the best-fit principle to achieve dynamic consolidation of the resources in the data center. Finally, experimental results show that compared with state-of-the-art algorithms under different conditions, the proposed method reduces energy consumption by approximately 18.25%, VM execution costs by approximately 21.34%, and service level agreement (SLA) violations by approximately 90.51%.

Poker is the typical game of incomplete information, and remains a longstanding challenge problem in artificial intelligence (AI). The poker game of Dou Dizhu has been viewed as a thorny topic in AI because of its own characteristics. This article introduces a developed Monte Carlo tree search (MCTS) method for Dou Dizhu to solve the decision making effectively. We built the winning rate prediction model (WRPM) to predict the winning rate of moves as the initial situation estimation and improve the model to be more applicable to different player roles. Then, the WRPM is embedded as the core algorithm into MCTS for extension and simulation and named it WRPM-MCTS. In addition, we also train a card distribution prediction model to predict the holding cards of opponents for further improving the performance of WRPM-MCTS on the agent of Dou Dizhu. Experiments show that the WRPM-MCTS has a statistically significant performance better than the pure MCTS and the pure WRPM. In the game with human players from an online game platform, the WRPM-MCTS-based agent had the winning rate of 52.86% in 4 000 000 games and ranked in top 1.22% among 500 000 human players, indicating that this agent had reached the expert level of humans.

With the development of cloud computing, an increasing number of applications in different fields have been deployed to the cloud. In this process, the real-time scheduling of multiple workflows composed of tasks from these different applications must consider various influencing factors that strongly affect scheduling performance. This paper proposes a real-time multiple-workflow scheduling (RMWS) scheme to schedule workflows dynamically with minimum cost under different deadline constraints. Due to the uncertainty of workflow arrival time and specification, RMWS dynamically allocates tasks and divides the scheduling process into three stages. First, when a new workflow arrives, the latest start time and the latest finish time of each task are calculated according to the deadline, and the subdeadline of each task is obtained by probabilistic upward ranking. Then, each ready task is allocated according to its subdeadline and the increased cost of the virtual machine (VM). Meanwhile, only one waiting task can be assigned to each VM to reduce delay fluctuations. Finally, when the task is completed on the assigned VM, all the parameters of the relevant tasks are updated before allocating them to appropriate VMs. The experimental results based on four real-world workflow traces show that the proposed algorithm is superior to two state-of-the-art algorithms in terms of total rental cost, resource utilization, success rate and deadline deviation under different conditions.

Weakly supervised semantic segmentation under image-level annotations is effectiveness for real-world applications. The small and sparse discriminative regions obtained from an image classification network that are typically used as the important initial location of semantic segmentation also form the bottleneck. Although deep convolutional neural networks (DCNNs) have exhibited promising performances for single-label image classification tasks, images of the real-world usually contain multiple categories, which is still an open problem. So, the problem of obtaining high-confidence discriminative regions from multi-label classification networks remains unsolved. To solve this problem, this article proposes an innovative three-step framework within the perspective of multi-object proposal generation. First, an image is divided into candidate boxes using the object proposal method. The candidate boxes are sent to a single-classification network to obtain the discriminative regions. Second, the discriminative regions are aggregated to obtain a high-confidence seed map. Third, the seed cues grow on the feature maps of high-level semantics produced by a backbone segmentation network. Experiments are carried out on the PASCAL VOC 2012 dataset to verify the effectiveness of our approach, which is shown to outperform other baseline image segmentation methods.

Summary Mobile ad hoc Network (MANET) is a cluster of moveable devices connected through a wireless medium to design network with rapidly changing topologies due to mobility. MANETs are applicable in variety of innovative application scenarios where smart devices exchange data among each other. In this case, security of data is the major concern to provide dependable solution to users. This article presents a secure mechanism for data transfer where sender splits the data into fragments and receiver gets the actual data by assimilating the data fragments. We have presented an Enhanced Secured Lempel‐Ziv‐Welch (ES‐LZW) algorithm that provides cryptographic operations for secure data transfer. In proposed model, we have utilized the disjoint paths to transfer the data fragments from sender side and assimilate these fragments at receiver to get the original data. The messages containing data fragments are compressed and encrypted as well. Our scheme ensures confidentiality, integrity, efficient memory utilization, and resilience against node compromising attacks. We have validated our work through extensive simulations in NS‐2.35 using TCL and C language. Results prove that our scheme reduces memory consumption along with less encryption and decryption cost as compared to blowfish especially when plaintext has more repetitive data. We have also analyzed the impact of creating data fragments, fraction of communication compromised, and probability to compromise the data fragments by subverting intermediaries.

This paper briefly introduce a method to evaluate the reliability of a cloud rendering system by using probability models. An extended discrete-time Markov chain (DTMC) is proposed considering the QoS (Quality of Service). Then, some properties defined from 3 aspects give full consideration to the processes of rendering tasks, which can be verified by performing PRISM in a quantitative way. Finally, the experimental results demonstrate that our method can ensure and improve the QoS reliability of the cloud rendering system.

This paper proposes an approach to evaluate the reliability of cloud rendering system. After the requirement analysis, the rendering system was divided into three modules: preparing files, requesting resources, and rendering task execution. Each module may have an exception that will reduce reliability, and has the ability to recover it. To expose these details, the discrete-time Markov chain (DTMC) is improved to formalize the cloud rendering system. The model contains an abnormal state set representing exceptions and errors such as file corruption and failure to rendering subtasks. Then, a series of formal properties are defined to describe reliability in detail. The proposed method gives full consideration to the processes of rendering tasks. Finally, the properties are verified by performing PRISM in a quantitative way. The experiment shows that our method is effective to evaluate the reliability of the cloud rendering system.

There are huge amount of tasks and data to be processed in cloud rendering environment.How to effectively schedule them is the key to ensure the overall performance of the cloud rendering environment.In this paper, an improved task scheduling algorithm based on genetic algorithm (PMGA) and probability model is proposed, which aims to minimize the total time and cost of task scheduling .First, the fitness function relating to the total time and task cost is improved under the consideration of the user's satisfaction to the rendering services.Then, a probability model is constructed for the scheduling algorithm, which is used to achieve the non-linear adaptive adjustment of the crossover rate function and the mutation rate function.As a result, the evolutionary ability of poor individuals in the population can be enhanced, avoiding the stagnation in the early stages.Finally, experiments are performed to demonstrate that PMGA has a better ability of optimization than that of traditional adaptive genetic algorithm (AGA).Our approach contributes to reduce the total time and the scheduling cost of the cloud rendering tasks.

In open and changeful Internet, the enterprise business process needs to be organized or restructured dynamically in order to adapt to environment changes and business logic updates. The solution of Web service and service-oriented architecture (SOA) provides a promising approach. The business processes working as a temporary workflow can be composed by distributed services. However, the cross-organizational service feature of business process requires considering not only the functional requirements but also the timed constraints. The timed property plays an important role in service interactions between business processes, such as timed activity, timeout and timed deadlock. Thus, if time requirements cannot be guaranteed, the new created business process will not be acceptable. In this paper, it proposes a framework of using Petri Net to model timed service business process. First, it defines the behavior model of service business process and gives process composition patterns for different structural forms. Second, service model is extended with time specifications, describing timed constraints among business activity interactions. Third, to support further verifications, it introduces a method for the automatic timed properties generation in the form of temporal logic formulae. Our framework gives a reference in practice to formalize service business process into timed service model.