The goal of this paper is to manipulate image attributes using text description. Although many methods can synthesize images with new properties from text, they cannot fully preserve the text-independent content of the original image. There are two major limitations: (1) Some important details in image sub-region will be lost in the process of image modification; (2) Compared with the original image, the shape and edge of the object in the modified image will be more blurred. Therefore, we propose a novel framework edge aware generative adversarial network (EA-GAN) that uses edge information to guide image modification, which ensures the network’s ability to identify local regions and realizes accurate modification of image sub-regions. At the same time, an edge reconstruction loss (ERLoss) is added to the generator to constrain the generation of edges, generate sharper edges, and improve the clarity of the image. The experimental data on the CUB and Oxford-102 datasets show that the algorithm used in this paper can well distinguish the corresponding image features in the conditional text, and modify the image attribute of specific regions in the image.
Network Function Virtualization (NFV) has becoming an emerging technology for ensuring the reliability and security of data flows. The virtual network function (VNF) embedding problem, which tries to minimize the embedding cost has attracted extensive interests recently. However, the existing works always assume the fixed execution order of VNFs, which limits their application. Thus, we investigate the VNF embedding problem without such limitations in this paper. Firstly, a general transformation framework is proposed for the NFV-enabled unicast routing with adjustable order. Then, an optimal algorithm is proposed for the unicast VNF embedding without delay constraints. Additionally, an efficient algorithm is also proposed for such problem with delay constraints. Finally, we evaluate the proposed algorithms via numerical results, which show that our algorithms significantly outperform the existing benchmarks.
The lower limb exoskeleton robot system is one of the significant tools for the rehabilitation of patients with knee arthritis, which helps to enhance the health of patients and upgrade their quality of life. However, the unexplained gait recognition model decreases the prediction accuracy of the exoskeleton system. The existing explainable models are seldom used in the domain of gait recognition due to their high complexity and large computation. To strengthen the transparency of the model, SHapley Additive exPlanations (SHAP) is applied to gait recognition for the first time in this paper, and an interpretable model framework that can be applied to any lower limb exoskeleton is proposed. Compared with the existing methods, SHAP has a more solid theoretical basis and more efficient calculation methods. The proposed framework can find the relationship between input features and gait prediction, to identify the optimal sensor combination. Additionally, The structure of the gait recognition model can be optimized by adjusting the feature attention of the model with the feature crossover method, and the accuracy of the model can be upgraded by more than 7.12% on average.
The time-limited order is a new type of real-time delivery service that platforms need to complete the order delivery with different time granularity (e.g. one day, two days, or three days). Predicting the number of time-limited orders plays an important role for real-time order delivery allocation and anomaly detection in logistics IoT scenarios. However, the impact between orders of different time granularity is complex and the contribution of the static order features is unknown. Previous order predicting methods are not suitable for time-limited orders because they do not fully consider the dependencies among orders with different time granularity. In this paper, we propose a spatial-temporal framework based on stacked long short-term memory networks (LSTM) and deep & cross network (DCN) to take the dependencies of multi-scale temporal features into account and fuse cross-domain static features effectively. In addition, we utilize a multi-head attention mechanism to model the heterogeneous strengths of different dependencies. We evaluate our model on a real-world dataset with about 400,000 orders from one of the largest logistics companies in China. The evaluation results show that the Mean Absolute Error and R2 score of our method achieve 9.407 and 0.948, outperforming state-of-the-art solutions.
Mobile social networks facilitate people's communication and exchange. However, the dissemination of rumor in mobile social networks is not conducive to building a harmonious network environment. Studying the dissemination mechanism of rumor in mobile social networks can reduce the harm caused by rumor to the network environment. Hence, we propose a new model of rumor dissemination in mobile social networks considering the difference of user's reaction when receiving information and the difference of user's behavior when processing information. Then, we make a stability analysis of the model. Finally, we conduct simulation experiments on the dissemination process of rumor in mobile social networks. The simulation results show that our proposed rumor dissemination model can effectively predict the dissemination trend of rumor in mobile social networks. Therefore, our work will contribute to the study of the dissemination mechanism of rumor in mobile social networks.
With the advent of the “Internet+” era, the IoT has developed rapidly and is gradually penetrating into all fields of life. While the scale of IoT devices is showing an explosive growth trend, the importance of IoT security is also becoming more and more prominent in the rapid development of the IoT. In order to assist the identification of Internet-connected devices and further identify the vulnerability information of devices to achieve security protection for IoT devices, we construct a large-scale, diverse, and high-coverage dataset of IoT devices by automated and semi-automated means. In this dataset, each piece of data contains the information of the category, brand, and product model of the device. The large-scale, diverse, and high-coverage properties of the dataset are fully validated through our statistical analysis and experimental applications.
Fog computing, which provides low-latency computing services at the network edge, is an enabler for the emerging Internet of Things (IoT) systems. However, due to the limited capacity of the fog devices, a large number of tasks will still be offloaded to the cloud for processing. This can overload the backbone network and cause excessive delays. In addition, the distribution of IoT devices is uneven. This paper groups fog nodes into a fog federation to increase the number of tasks that fog nodes can handle. The boom in fog computing has made fog pricing an important issue. But so far, no one has set a price for the federation. This article uses the Stackelberg model to simulate the interaction process between fog federation and IoT devices. First, fog devices bundle certain computing power, memory, and some other resources into a computing resource block (CRB). Then, fog devices act as a leader to release the price of CRB, and IoT devices act as a follower to determine the amount of CRB purchased. This article introduces how IoT can buy the right amount of CRB to minimize its cost and how CRB is priced to maximize the revenue of the fog federation.
The explosion of traffic caused by the rapid growth of multimedia services of Internet of Vehicles (IoV) has brought heavy load to mobile networks. The edge caching of the Internet of vehicles is considered as a promising technology. When the existing content caching strategy is used in the vehicle network, it faces the challenge of high content caching delay caused by the high-speed mobility of vehicle users and insufficient social relations. To address these challenges, this paper proposes a Cooperative Edge Caching Scheme based on Mobility Prediction and Society Aware (CCMPSA). In this strategy, the Long Short-Term Memory (LSTM) network is used to predict the location of the vehicle at the next moment, the vehicle cache nodes are selected according to the social relationship reflected by the similarity of interest and communication probability between the vehicles, and the dynamic decision of the content cache problem is realized by deep reinforcement learning. The simulation results show that the performance of the proposed strategy is better than random caching and non-cooperative caching algorithms, and it not only reduces the content transmission delay and improves the cache hit ratio, but also improves the experience quality of the whole system.
Liquid identification is an essential technology for water safety monitoring. This paper shows the feasibility of identifying liquid using millimeter wave (mmWave) signals. The inherent principle comes from that the fine-grained mmWave signals can capture signal attenuation, phase shift, and propagation delay when penetrating the liquid. We have conducted a preliminary experiment to prove the effectiveness of using mmWave for liquid identification. However, after moving the container, the identification accuracy will drop significantly. To address this challenge, we propose a robust mmWave-based liquid identification approach MmLiquid, which uses a container position information filtering (CPIF) scheme to eliminate the influence of different container positions. MmLiquid will extract container position-independent information from the original mmWave signals and train a deep complex model (DCN) for accurate liquid identification. To further improve the identification performance, we set up an identification environment with two reflective surfaces to capture effective mmWave signals that contain more liquids information. We implement MmLiquid using commercial mmWave devices. Experimental results on 16 kinds of liquids at 24 different container positions show that MmLiquid can achieve an average liquid identification accuracy of 97.6%.
The industrial Internet of Things (IIoT) can provide production management services for customers, but also faces cyber-attacks and security issues. The network security situation prediction describes the security status of the network from a holistic perspective for the coming period. Existing network security situation prediction models suffer from low prediction accuracy and do not apply to industrial IoT scenarios. To address the above problems, an ISSA-BiLSTM-based security situation prediction model for industrial IoT is proposed. Firstly, a situational assessment index system is proposed for the characteristics of industrial IoT, and the real situational values of the used industrial IoT datasets are calculated based on this index system. Besides, the opposition-based learning and Lévy flight strategy are introduced to improve the sparrow search algorithm (SSA) to avoid it from falling into local optimum. At last, we use the improved SSA (ISSA) to search for optimality of the relevant parameters of the BiLSTM model to ensure the accuracy of the prediction. With simulation experiments, we verify that the ISSA-BiLSTM model has smaller errors and higher fitness.
In recent years, the wide spread adoption of mobile applications exerts a great burden on backhaul links. Mobile edge computing (MEC) alleviates the problem by enabling mobile edge devices with cache storage to reduce network congestion and content delivery latency. In this paper, we study the problem of bandwidth minimization under Age of Information (AoI) constraint in mobile edge caching system with heterogeneous devices. We present a content refreshing mechanism, and propose an algorithm extended from the classical genetic algorithm to minimize average AoI. Specially, mobile users declare requests of the content service provision that are dynamically collected at base station (BS). Upon receiving these requests, the BS can decide how to allocate the bandwidth for updating the content, so that the average AoI in the network is minimized. Unlike existing work, this paper also takes into account the diverse QoS requirements of heterogeneous nodes in data transmission. Extensive simulations are conducted to validate the analytical results. Compared to other conventional algorithms, the results indicate that the proposed scheme can effectively reduce the average AoI.
As a scientific research method to reveal the intrinsic functional properties of complex network systems, Community Detection has already become one of the most popular research topics in complex networks. The typical label propagation algorithms are very suitable for large-scale networks due to their approximate linear time complexity. But too many random strategies in the algorithms make it not stable enough. For that reason, this paper proposes a Community Detection Algorithm Fusing Node Similarity and Label Propagation (FNSLP). First, the algorithm preprocesses the neighboring nodes of the seed nodes by node similarity to reduce the kinds of the initial label. Combined with nodes’ influence, the label propagation ability is calculated. Then, the label selection of nodes is assisted by an improved label update strategy, which reduces the phenomenon of label oscillation and improves the accuracy and stability of label selection. Experimental results show that in four real networks, the algorithm achieves the maximum Modularity value on 75% of the datasets. In multiple artificial benchmark networks with different mixing parameters, the algorithm's Normalized Mutual Information value reaches the maximum value.
IoT binary similarity detection is a way to determine whether two IoT components have a homology relationship. It is used to address security concerns arising from the reuse of open source components in the IoT software supply chain. In order to solve the problems caused by different architectures and different optimization levels during compilation, we propose a graph neural network based similarity assessment model for IoT binary components. We introduce an attention mechanism to get graph-level embedding based on GraphSAGE node-level embedding extraction, witch considers the importance of function nodes. Through comparative experiments with other models, this model shows better performance with 96.96% accuracy.
Problems with parking have resulted in traffic congestion, social phobia, and smog, as well as an inefficient allocation of resources as a result of the city's growing population. The importance of computer-aided methods and existing methods for parking prediction are analyzed. On this basis, the existing deep learning and mobile deep learning methods were investigated in depth and found that the existing methods are dependent on the cloud server and also have issues related to accuracy and response time. MDLpark, a novel mobile deep learning architecture-based approach for parking occupancy prediction is proposed. The method is based on Temporal Convolutional Network (TCN), which uses a one-dimensional fully convolutional network architecture through TCN, which utilizes its convolution and dilation to make it dynamically adaptive to the prediction window and prediction response. We restructured the residual block of TCN by replacing the weight normalization with batch normalization, which is more stable than weight normalization. The proposed MDLpark prediction method is implemented based on Keras and TensorFlow, and a mobile deep learning parking application is designed. Experimental results show that compared with other models (TCN, LSTM, GRU and MLP), MDLpark achieves higher prediction accuracy with 97.6% accuracy (97.3%, 84.3%, 95.6%, and 95.7% for TCN, LSTM, GRU, and MLP, respectively). At the same time, through the developed model, the traveler's time to find a parking space is reduced, and the function of direction to these parking lots is provided.
In the process of individuals acquiring and sharing knowledge in online social networks, the difference of knowledge internalization ability, the lack of trust and incentive mechanism hinder the effective dissemination and prediction of knowledge. Therefore, it is significant to find effective ways to predict and promote knowledge dissemination in online social networks. In this paper, we establish a novel dynamics model, which considers the complex psychological cognition and behavior of individuals, and adds two new states to describe the dynamic process of knowledge dissemination more accurately compared with the classical infectious disease model. Besides, we investigate the trend of knowledge dissemination and the stability of the proposed model. Our theoretical analysis shows the proposed model can effectively judge and predict the trend of knowledge dissemination through a threshold, and simulation experiments verify the proposed knowledge dissemination dynamics model is reasonable, and it can effectively promote knowledge dissemination.
Due to restricted resources such as computing power and battery capacity of Industrial Internet of Things (IIoT) equipments, computation-intensive tasks need to be migrated to edge or cloud servers for execution. To improve the processing efficiency of tasks with limited computation and network resources, we study the problem of joint allocation of network and computational resources in the cloud-edge collaborative IIoT, with the goal of minimizing the average task delay and total system energy consumption. To address this issue, we propose a prioritized action sampling-based Dueling DQN (PASD) algorithm to determine task offloading and resource allocation strategies. Finally, we evaluate PASD through large-scale simulation experiments and NBUFlow, which is an IoT experimental platform equipped with object recognition and pose detection applications. Compared with baselines, PASD has significant advantages in reducing the total energy consumption of the system, and has a good performance in reducing task delay and task throw rate.
The problem of anomaly detection in marine Argo data is studied. Based on the common and widely used DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm for anomaly detection in Argo data, as in other fields, there is a major problem in the application of DBSCAN algorithm, which is how to choose the appropriate parameter pairs. To solve this problem, this paper proposes an improved version of DBSCAN, namely CCMD-DBSCAN (DBSCAN based on the classification characteristics of marine data), which solves the above problem by studying the characteristics and laws of Argo data and is successfully applied to anomaly detection of marine data. The experimental results show that the new algorithm can not only determine the appropriate parameter pairs but also has a good anomaly detection effect.
IEEE 802.11ah is a wireless network protocol designed for large-scale Industrial Internet of Things (IIoT) scenarios. The restricted access window (RAW) mechanism introduced by IEEE 802.11ah assigns the nodes into different RAW groups to reduce conflicts between nodes. The original RAW mechanism randomly divides nodes into RAW groups with the same duration, which cannot meet the different requirements of different nodes. In order to reasonably divide the nodes into different RAW groups, a RAW grouping algorithm based on outage probability (OP-RAW) is proposed in this paper. Outage probability is introduced to evaluate the quality of data transmission in IIoT, and the influencing factors is analyzed. OP-RAW calculates the time slot duration of the nodes according to the load and outage probability of different nodes, and then groups the nodes with the same time slot duration requirements into the same RAW group. The simulation results show that, compared with the traditional RAW grouping algorithm and other RAW group optimization algorithm, the OP-RAW can increase the throughput by about 15% and effectively reduce the transmission delay.
The Automatic Guided Vehicle (AGV) sorting system is a new way of sorting. Path planning is the key part to improve sorting efficiency of the system. In the existing methods, the multiflow network path planning method obtains optimal paths by transforming path planning problems into integer linear programming problems and solving them. However, due to the huge amount of calculation, it is hard to ensure the real-time performance and use the method in practice. In this paper, a Calculation Time Prediction-based Multiflow Network path planning method (CTPMN) is proposed to deal with the problem. Firstly, a new operating model is proposed, in which the time window of the AGV moving can be set according to the predicted calculation time of the path planning method. And then a support vector machine-based algorithm is used to learn the predicted calculation time based on the size of sorting area, number of sorting tasks and the maximum task distance. In addition, a task decomposition strategy is put forward to eliminate the constraint that the destination of any two tasks in the multi-flow network method cannot be the same. Finally, the effectiveness of the proposed method is verified by simulation experiments.
Due to the high mobility and limited transmission range of vehicles, the data download capacity of single vehicle is greatly limited, which brings poor performance to users. In this paper, we divide the data into blocks. We want to design a data block broadcasting scheme so that all vehicles can receive data blocks as many as possible in a base station(BS) range. We first give the mathematical model and find it is difficult to be solved directly. Then we design a heuristic algorithm for solving the problem. The main idea of our algorithm is to give each data block a weight. The data block with the largest weight is broadcast by BS, and several vehicles are selected to broadcast the remaining data blocks. We call our algorithm as the Iterative Strategy for Data Allocation(ISDA) algorithm. Then considering the actual situation, we subsequently propose the online algorithm. Through experiments and simulations, we prove that our scheme can effectively improve the data download rate and reduce the download delay.
