ABSTRACT Large‐scale model training in distributed data centers plays a crucial role in deep learning. Still, it faces significant challenges, including resource fragmentation, low bandwidth utilization, and complex task flow management. The problem is exacerbated by high‐speed, high‐capacity parameter synchronization, often exceeding several hundred Gbps, which leads to reduced throughput and computational inefficiencies. To address these challenges, this paper proposes an innovative approach that combines data parallelism, model parallelism, and dynamic load balancing. By integrating a Gray Markov Chain (GMC) and Markov Decision Process (MDP) model, the approach dynamically schedules resources and balances computational loads. The GMC model is used to predict future node loads, facilitating optimal weight matrix decomposition, while the MDP model adjusts data transmission paths to optimize network traffic management. The combination of these two models enhances both resource allocation and data flow optimization. Experimental results demonstrate that this integrated approach significantly improves throughput, resource utilization, and computational efficiency compared to traditional methods. The findings suggest that this hybrid approach performs exceptionally well in optimizing large‐scale distributed training tasks in multidata‐center environments, significantly improving the scalability and performance of deep learning workloads. This research shows promising implications for enhancing the efficiency and effectiveness of distributed training systems in high‐demand applications.

Identifying influential nodes in complex networks poses a significant challenge, requiring a delicate balance among efficiency, accuracy, and redundancy elimination. To address this, we propose TIG-IM (A Two-stage Inductive GNN for Influence Maximization), a novel framework based on GNN. In Stage 1, we construct a multi-task learning-based global scorer. This scorer utilizes a Graph Attention Network (GAT) to simultaneously regress the SIR propagation score and our newly proposed 3D-Bridging Centrality. This multi-task design enables efficient identification of cross-community bridge nodes to generate a high-quality candidate pool. In Stage 2, we design a local selector that introduces a composite loss function with sparsity and diversity regularization. This selector performs a secondary re ranking on the candidate subgraph to produce a final seed set that is both structurally dispersed and possesses complementary propagation potential. Theoretical analysis proves that our scorer satisfies monotonicity and that the local loss can be interpreted as a MAP estimation. Extensive experiments on twelve real-world networks from diverse domains, including social, academic, and biological, demonstrate that TIG-IM significantly outperforms various state-of-the-art baselines in both spreading effectiveness and computational efficiency. Ablation studies further validate the synergistic value of our three core components: 3D-Bridging Centrality, sparsity penalty, and diversity-based re-ranking.

Nowadays, unmanned aerial vehicle (UAV) ad hoc networks (UANETs) find application in a wide range of scenarios. Traditional routing protocols from mobile ad hoc networks (MANETs) or vehicular ad hoc networks (VANETs) cannot be directly deployed for UANETs. Therefore, we propose a multi-parameter metric routing protocol called technique for order preference by similarity to an ideal solution aided ad hoc on-demand distance vector routing protocol (TAAODV). During the route discovery phase, it utilizes the stability of neighbor nodes to determine the forwarding mechanism, thereby suppressing flooding. In the route selection phase, it uses newly defined accumulated pheromones as the basis for selecting the optimal path. These improvements in both phases address the issues of network storms and path stability in highly dynamic scenarios.We compared the TAAODV protocol with four classic routing protocols through simulation. The simulation outcomes reveal that the TAAODV routing approach offers substantial advantages in the packet delivery ratio (PDR), average end-to-end delay, MAC-layer routing overhead and the longevity of the network. Furthermore, TAAODV also performs well in critical metrics such as routing overhead and throughput when the packet sending rate increases.

The vulnerability of cryptographic keys to exposure represents a critical risk to modern cryptographic techniques. In this paper, we present an approach to enhancing the security of encrypted communication systems through the strategic incorporation of redundancy design. As hardware costs continue to decline, deploying such redundancy becomes increasingly feasible. This approach allows for the protection of individual keys through the amalgamation of multiple redundant keys, significantly lowering the probability of security breaches. The study's findings demonstrate that the proposed scheme not only enhances the resilience of encrypted communication systems against potential eavesdropping and key exposure but also provides a scalable solution for improving overall cryptographic security.

Precipitable water vapor (PWV), as an important representative parameter of atmospheric water vapor contents, can be obtained by means of Global Navigation Satellite Systems (GNSS) using both ground-based and space-borne observation techniques. However, the PWV prediction models currently accessible tend to be simplistic combinations or individual models. In this study, we develop a WOA-CNN-LSTM-Attention model to predict PWV, which takes the sixteen GNSS PWV values near the HKKP station as characteristic parameters and the spatial relationship between the point of interest and its neighboring GNSS stations into consideration. An optimal model via the whale optimization algorithm (WOA) is investigated by using a wavelet analysis to separate noises, through combining convolutional neural network (CNN), long short-term memory neural network (LSTM) and attention mechanism. Results show that considerable improvement in the prediction accuracy has been achieved through a comparison between CNN-LSTM-Attention and the conventional LSTM and CNN-LSTM models. In terms of long-term predictability, CNN-LSTM-Attention is proven to be a superior model when 8 features are incorporated. The model’s root mean square error (RMSE) is 2.30 mm which is reduced by 20.42 % than in the case of 0 feature is used. As a further analysis, we also examine the prediction performance of various models for hourly PWV using 7, 15, 30, 60 and 90 days of data as different lengths of training. The results show that CNN-LSTM-Attention has a better prediction effect when the training length is 30 days, the RMSE is 0.74 mm and the Nash-Sutcliffe efficiency coefficient (NSE) is 0.98.

It is extremely important for autonomous vehicles to automatically complete refueling and various other tasks. However, most of the existing approaches are only suitable for conventional vehicles with large fuel tank, while the routing performance degrades, especially for electric vehicles and aerial vehicles that are constrained by small fuel capacity. This paper proposes a fuel-aware routing framework for autonomous vehicle that is employed for uninterruptedly visiting targets in logistics industry. By decoupling the topology representation and optimization process, the framework can express various optimization objectives, such as fuel consumption, trip delay and refueling cost, in a consistent way. For the fixed-path refueling problem, we propose an iterative method to improve the routing efficiency by minimizing unnecessary fuel reservation and design an efficient online solution to make trade off between routing efficiency and computation complexity. For the tour refueling problem, we propose a two-stage approach, which can simplify the routing calculation based on the solution to the fixed-path refueling problem. The simulation results demonstrate that the proposed scheme outperforms the alternative approaches in terms of both routing efficiency and computation complexity, while supporting for joint optimization of multiple objectives.

Data compression is extremely important for all kinds of networking applications, but most existing compression algorithms are only based on utilizing the redundancy characteristics of data source. This paper proposes a novel compression paradigm, which explores another kind of redundancy that is provided by the correlated state information shared between sender and receiver of a local transmission. Such kind of redundancy reflects the knowledge of two peers about each other before a transmission really takes place, with the aid of possibly existing cloud data center or sensing interfaces equipped. We formulate a typical representation of shared context as inequality constraints for the general case in n-dimensional Euclidean space, and provide a constructive proof for the existence of a bijective mapping used for compression and decompression with the shared context. By experimenting on compressing geographic spatial-temporal data for efficient transmissions, analysis and simulation demonstrate that the proposed scheme outperforms the well-known Huffman coding and Delta algorithms in terms of compression ratio.

Synchronization among screens is very important for correctly working of tiled display systems, which is regarded as one of fundamental technologies for providing a very large and high resolution display. This paper proposes a novel approach to synchronization of animation and video-embedded 3D content, so that complicated 3d virtual scenarios can be playbacked in a tiled fashion over multiple devices. Based on a grid configuration, a display controller collects the information of rendering threads and sends synchronization beacons to all tiled windows in a multicast way and each tiled window will adapt its render progress accordingly. A ready bar-based synchronization method for video is used to overcome the influence of packet losses and the variance of buffering and networking delay on video frames. The experiment results show that all kinds of 3D contents, including 3D models, textures, animations, virtual cursors and videos embedded can be rendered in a consistent way even under poor networking conditions.

Vehicular ad hoc networks (VANETs) are viewed as an effective way to disseminate content among vehicles on the road. While most of the current research efforts in VANETs focus on improving the packet delivery performance, only limited work aims to provide cost-efficient solutions for content downloading. This paper proposes a novel approach for the vehicles to download a common content in a cost-efficient way. The basic idea is to stimulate the vehicles to download the content cooperatively in mutually disjoint coalitions. To study the cooperation among the vehicles, a coalitional game model is proposed. Moreover, a distributed coalition formation algorithm is designed to obtain a stable coalition structure and minimize the total communication cost. Numerical results show that the proposed content download approach can achieve a considerable communication cost reduction compared with the non-cooperative case.

Mobile ad hoc networks (MANETs) are designed for "a special purpose'', usually supporting some special functions, such as sensing environments, providing services. However, the traditional model of service accessing, such as the server/client model, can not be directly applied to MANETs, as the node, running the server, providing services, may be multi-hop distant, even mobile and thus unreachable from the client. This paper proposes a novel approach, by which a node can query other nodes by using the standard set of SQL (Structured Query Language) functions, which is the foundation of mobile databases. Analysis and simulation demonstrate that there is a tradeoff between the communication overhead and the query accuracy, which should be carefully weighed in the real network environment.

In order to combat the exposed terminal problem and the interference accumulation phenomenon in wireless channels, this paper proposed a new Asymmetrical Dual Busy Tones (ADBT) MAC protocol. By introducing BT-CTS/BTACK mechanism and the approximate algorithm of interfering nodes, ADBT MAC can solve the exposed terminal problem and the hidden terminal problem in the environment with large interference range. Moreover, the effect of the interference accumulation phenomenon can be alleviated by the power adjustment approach for the busy tones. The network performances of DCF, DBTMA, VPDBT and ADBT MAC protocols are compared via OPNET-based simulation experiments, which verify that ADBT is able to prevent packet collisions effectively and improve the channel utilization significantly.

Abstract IP autoconfiguration of the mobile node addresses is important in the practical usage of most mobilead hocnetworks (MANETs). This paper proposes domain‐based autoconfiguration framework (DACF), a novel approach for the efficient address autoconfiguration of MANETs. To construct a hierarchy of addresses, this framework defines a loose domain structure where nodes in the same domain may roam to different locations after they are configured. This framework uses the passive Duplicate Address Detection(PDAD) but the proposed domain structure is able to reduce the initial conflict probability and accelerate the conflict resolution significantly. To evaluate the correctness and efficiency of the proposed framework, we also present an exemplified full‐functioned implementation of the proposed framework. Through the detailed analysis and simulation, we believe the proposed scheme provides a promising autoconfiguration framework for large‐scale MANETs. Copyright © 2008 John Wiley & Sons, Ltd.