In recent years, unmanned aerial vehicles (UAVs) equipped with wireless chargers have been utilized as mobile charging units to flexibly recharge energy-depleted sensor nodes in wireless sensor networks. However, the endurance of UAVs remains a critical challenge. By deploying wireless charging pads to provide energy replenishment, UAVs can sustain their operations and successfully complete their charging tasks. Therefore, a key problem is how to effectively deploy the minimum number of wireless charging pads while establishing at least one feasible charging path from the base station (BS). This ensures that the UAV can reach and recharge all sensor nodes from BS. When designing an optimization algorithm for wireless charging pad deployment, connectivity, coverage, and geometric properties must be considered simultaneously. However, previous approaches often employed greedy algorithms to solve the optimal deployment problem, treating coverage and connectivity as interdependent properties. This led to excessive constraints on the placement of wireless charging pads, as each newly added charging pad had to satisfy both properties at the same time. Additionally, previous works overlooked a critical issue, which is avoiding the occurrence of isolated sensor nodes in uncovered fragmented regions, in deployment. Failing to address this issue would require additional deployment costs to compensate for uncovered nodes. This study focuses on addressing these challenges by proposing a novel deployment strategy to enhance the energy replenishment efficiency and overall performance of wireless charging sensor networks. To overcome the limitations of previous methods, we apply computational geometry techniques to develop a more effective deployment algorithm. Our proposed onion-peeling like wireless charging pad deployment algorithm deploys charging pads layer by layer from the outermost region inward, prioritizing coverage before connectivity. Simulation results demonstrate that the proposed approach significantly reduces the number of required wireless charging pads compared to existing methods.

Network motif is a kind of frequently occurring subgraph that reflects local topology in graphs. Although network motif has been studied in graph analytics, e.g., social network and biological network, it is yet unclear whether network motif is useful for analyzing online transaction network that is generated in applications such as instant messaging and e-commerce. In this work, we analyze online transaction networks from the perspective of network motif. We define vertex features based on size-2 and size-3 motifs, and introduce motif-based centrality measurements. We further design motif-based vertex embedding that integrates weighted motif counts and centrality measurements. Afterward, we implement a distributed framework for motif detection in large-scale online transaction networks. To understand the effectiveness of motif for analyzing online transaction network, we study the statistical distribution of motifs in various kinds of graphs in Tencent and assess the benefit of motif-based embedding in a range of downstream graph analytical tasks. Empirical results show that our proposed method can efficiently find motifs in large-scale graphs, help interpretability, and benefit downstream tasks.

Graph neural networks (GNNs) have achieved great success in many graph-based applications. However, the enormous size and high sparsity level of graphs hinder their applications under industrial scenarios. Although some scalable GNNs are proposed for large-scale graphs, they adopt a fixed K-hop neighborhood for each node, thus facing the over-smoothing issue when adopting large propagation depths for nodes within sparse regions. To tackle the above issue, we propose a new GNN architecture --- Graph Attention Multi-Layer Perceptron (GAMLP), which can capture the underlying correlations between different scales of graph knowledge. We have deployed GAMLP in Tencent with the Angel platform, and we further evaluate GAMLP on both real-world datasets and large-scale industrial datasets. Extensive experiments on these 14 graph datasets demonstrate that GAMLP achieves state-of-the-art performance while enjoying high scalability and efficiency. Specifically, it outperforms GAT by 1.3% regarding predictive accuracy on our large-scale Tencent Video dataset while achieving up to 50x training speedup. Besides, it ranks top-1 on both the leaderboards of the largest homogeneous and heterogeneous graph (i.e., ogbn-papers100M and ogbn-mag) of Open Graph Benchmark.

Graph neural networks (GNNs) have achieved state-of-the-art performance in various graph-based tasks. However, as mainstream GNNs are designed based on the neural message passing mechanism, they do not scale well to data size and message passing steps. Although there has been an emerging interest in the design of scalable GNNs, current researches focus on specific GNN design, rather than the general design space, limiting the discovery of potential scalable GNN models. This paper proposes PaSca, a new paradigm and system that offers a principled approach to systemically construct and explore the design space for scalable GNNs, rather than studying individual designs. Through deconstructing the message passing mechanism, PaSca presents a novel Scalable Graph Neural Architecture Paradigm (SGAP), together with a general architecture design space consisting of 150k different designs. Following the paradigm, we implement an auto-search engine that can automatically search well-performing and scalable GNN architectures to balance the trade-off between multiple criteria (e.g., accuracy and efficiency) via multi-objective optimization. Empirical studies on ten benchmark datasets demonstrate that the representative instances (i.e., PaSca-V1, V2, and V3) discovered by our system achieve consistent performance among competitive baselines. Concretely, PaSca-V3 outperforms the state-of-the-art GNN method JK-Net by 0.4% in terms of predictive accuracy on our large industry dataset while achieving up to 28.3 × training speedups.

Recent works reveal that feature or label smoothing lies at the core of Graph Neural Networks (GNNs). Concretely, they show feature smoothing combined with simple linear regression achieves comparable performance with the carefully designed GNNs, and a simple MLP model with label smoothing of its prediction can outperform the vanilla GCN. Though an interesting finding, smoothing has not been well understood, especially regarding how to control the extent of smoothness. Intuitively, too small or too large smoothing iterations may cause under-smoothing or over-smoothing and can lead to sub-optimal performance. Moreover, the extent of smoothness is node-specific, depending on its degree and local structure. To this end, we propose a novel algorithm called node-dependent local smoothing (NDLS), which aims to control the smoothness of every node by setting a node-specific smoothing iteration. Specifically, NDLS computes influence scores based on the adjacency matrix and selects the iteration number by setting a threshold on the scores. Once selected, the iteration number can be applied to both feature smoothing and label smoothing. Experimental results demonstrate that NDLS enjoys high accuracy -- state-of-the-art performance on node classifications tasks, flexibility -- can be incorporated with any models, scalability and efficiency -- can support large scale graphs with fast training.

The fast advancement of Wireless Sensor Network (WSN) technologies has initiated different perspectives for applications from traditional wireless networks and wire networks. For example, they can be set up in indoor environments which integrate varieties of sensors with network capabilities to provide indoor environment monitoring or personnel location monitoring. Zig Bee is a commonly used transmission technology of indoor positioning. However, the accuracy of Zig Bee positioning is usually far from satisfactory due to the strength and interference of signals. In this paper, we strive to improve the accuracy of Zig Bee positioning and implement an indoor personnel tracking system. Two methods, Neighbor Area Majority Vote Priority Correction and Environment Parameter Correction, are proposed to promote the accuracy of Zig Bee positioning. The experiment results prove that our methods can largely increase accuracy of Zig Bee positioning and provide useful personnel tracking technology.

Localization and object tracking are important issues in wireless sensor networks. Many previous studies assume that a fixed equation can relate the received signal strength indication (RSSI) to the distance between two nodes. However, this assumption isn't suitable for real world since the RSSI can be easily affected by many factors. Thus, we adopt a more practical assumption that the higher the RSSI becomes, the closer the two nodes are, and vice versa. In our work, two tracking algorithms are proposed for random anchor deployment scenarios. The first one is a hybrid method while in the second algorithm distributed stronger signal virtual target algorithm (SSVT), a novel method to find virtual target is provided to estimate the current object location. Simulation results show that SSVT is more accurate, stable, and adaptable than previous work, especially for sparse anchor deployment networks which are more cost-effective.

Wireless sensor nodes are commonly deployed in outdoor or hazardous environments. Due to limited resources and power consumption required to perform tasks, these nodes may experience power shortages and thus lead to the disconnection of the whole wireless sensor network. To prolong the lifetime of the network, new technologies are developed to wirelessly recharge the sensor nodes via mobile machines. Previous works have considered applying wireless charging to elevate the network lifetime without defining performance optimization. An interesting issue is the effective partitioning of the network for more than one charging machine to patrol over. In this work, we aim to design and analyse three network partition methods, namely the tier-based partition, the sector-based partition, and the mixed partition, for charging scheduling with mobile charging machines so that the resulting sub-networks exhibits approximately the same total energy consumption rate. To the best of our knowledge, no optimal partition method has been proposed before. Moreover, we show that the largest diameter of the partitioned areas from the mixed partition is shorter than those from the tier-based partition and the sector-based partition. This indicates that the delay time can be reduced when a mobile charging machine runs for a dying sensor using the mixed partition method.

Frequent pattern mining (FPM) is important in data mining field with Apriori algorithm to be one of the commonly used approaches to solve it. However, Aprori algorithm encounters an issue that the computation time increases dramatically when data size increases and when the threshold is small. Many parallel algorithms have been proposed to speed up the computation using computer clusters or grid systems. GPUs have also been applied on FPM with only few adopting OpenCL although OpenCL has the advantage of being platform independent. Thus, the aim of this research is to develop efficient parallel Aprori strategy using GPU and OpenCL. Our novel method, Candidate Slicing Frequent Pattern Mining (CSFPM) algorithm, improves over the previous method by slicing candidate information to better balance the load between processing units. This strategy is proved to be more efficient according to our experiments. For example, CSFPM is at most 2.6 times faster than the previous method. Therefore, CSFPM is an efficient parallel Apriori algorithm which can reduce computation time and improve overall performance.

For mobile sensor networks, part or all of the nodes are with moving capabilities. It's essential to find energy-efficient routing path for nodes to communicate with each other. Broadcast and multicast are important issues in mobile sensor networks and much work has been done in finding energy-efficient routing methods for the broadcasting and multicasting process. In previous methods, the possibility of having different query-receiving probability between nodes was not considered. We consider a novel problem, namely irregular multicasting problem with the situation when nodes in the network may have different probabilities of receiving queries from the base station. Then, a new protocol, called Distance Confined Multicast Routing (DCMR), is proposed to solve this routing problem efficiently. Simulation demonstrated that DCMR is more energy-efficient and energy-balanced than the previously proposed methods.

Besides sensing the environment variables, the application of localization in wireless sensor networks has became an important research subject. Unlike the other range-free localization schemes which are not effective in real time performance, we propose a real time algorithm, which determines the location of the moving object based on dynamically changing signal strength. The simulation results demonstrated our algorithm is more effective and precise in the sense of real time localization scheme compared with previous method.

Besides sensing the environment variables and detecting the events in the deployment area, tracking of objects has gained much attention in the wireless sensor network research fields. Unlike the other range-free localization schemes which are not effective in real time performance, we propose a mechanism based on the received signal strength to develop real time object-tracking strategy, which determines the location of the moving object according to dynamically changing signal strength. Moreover, to improve the tracking accuracy and to be more practical, we consider a comprehensive approach, called CAUSS (Comprehensive Algorithm Using Signaling Strength), which can be applied in deployment which may result in various anchor node coverage. Simulation results demonstrated that our algorithm is more effective in the sense of real time tracking scheme compared with previous results.

Due to the fast advancement of network technologies, the study of service-oriented architectures (SOA) has attracted much attention recently, taking advantage of the benefits of distributed computing and integration. Web service is a very popular and widely accepted implementation of SOA. The use of Web services holds the advantages of a loosely-coupled system where all components can be developed at independent platforms and be connected with the Web service protocols via the network. However, the tendency for methods of applying Web services over the network is getting more and more complex. Many applications rely on not just one Web service, but a whole school of them. Thus, how to compose and integrate different Web services efficiently to provide complicated network services has become an essential topic in system development and design. This paper proposes a new problem which investigates the possibility of minimizing the number of hops of Web services while trying to finish a set of tasks. It also provides a polynomial-time, optimal Web service integration method using greedy strategy to integrate the Web services in order to complete those tasks. This method can achieve the goal of using the minimum number of hops of Web services over the network.

This paper explores the tracers deployment problem for IP traceback methods how many and where the tracers should be deployed in the network to be effective for locating the attack origins. The minimizing the number of tracers deployment problems depended on locating the attack origins are defined. The problem is proved to be NP-complete. A heuristic method which can guarantee that the distance between any attack origin and its first met tracer be within an assigned distance is proposed. The upper bound for the probability of an undetected attack node can be calculated in advance and used to evaluate the number of tracers needed for the proposed heuristic method. Extended simulations are performed to study the performance of the tracers deployment.