Search

Establishing correct correspondences between two images should consider both local and global spatial context. Given putative correspondences of feature points in two views, in this paper, we propose Order-Aware Network, which infers the probabilities of correspondences being inliers and regresses the relative pose encoded by the essential or fundamental matrix. Specifically, this proposed network is built hierarchically and comprises three operations. First, to capture the local context of sparse correspondences, the network clusters unordered input correspondences by learning a soft assignment matrix. These clusters are in canonical order and invariant to input permutations. Next, the clusters are spatially correlated to encode the global context of correspondences. After that, the context-encoded clusters are interpolated back to the original size and position to build a hierarchical architecture. We intensively experiment on both outdoor and indoor datasets. The accuracy of the two-view geometry and correspondences are significantly improved over the state-of-the-arts. Besides, based on the proposed method and advanced local feature, we won the first place in CVPR 2019 image matching workshop challenge and also achieve state-of-the-art results in the Visual Localization benchmark. Code is available at https://github.com/zjhthu/OANet.

An elastic stream computing system is expected to process dynamic and volatile data streams with low latency and high throughput in timely manner. Effective management of stream application is considered one of the keys to achieve elastic computing by scaling in/out the workload of each computing node properly during runtime. Many existing work tried to build an elastic stream computing system from one perspective or at one level, which limited to some extent the system performance improvement. To address the problems brought by single level management, in this paper, we propose and implement a multi-level collaborative framework (called Mc-Stream) for elastic stream computing systems. This paper introduces our solution from the following aspects: (1) Extensive experiments show that system performance is affected by multiple factors locating at different levels. A multi-level collaborative optimization strategy can coordinate those factors and optimize the performance to a greater extent. (2) A system model is constructed to explain the multi-level collaborative framework, with the creation of topology model, data model and grouping model. The process of multi-level collaborative framework is formalized, including optimizing instances number, determining data stream load ratio among instances and deploying instances. (3) The system performance is optimized at multiple levels (user level, instance level, scheduling level, and resource level). It is further improved by the components of lightweight instances management, available resource-aware data stream redirection, fast and effective scheduling management, and asynchronous runtime redeployment without state loss. (4) Mc-Stream is implemented on top of Apache Storm platform. Metrics are evaluated with real-world stream applications, such as the fulfillment of system latency, throughput and resources utilization. Experimental results show the significant improvements made by Mc-Stream: reducing average system latency by 32%, increasing average system throughput by 26% and average resources utilization by 34%, compared with existing state-of-the-art scheduling strategies.

Cyber-physical systems (CPSs) are a class of systems integrating cyber and physical components, and their security issues have gained a lot of attention in recent years. CPSs are modeled as hybrid systems in this letter since the logical and physical behaviors of CPS can be mapped to the discrete-state and continuous-state dynamics of the hybrid system, respectively. Motivated by the importance of situation awareness in an adversarial environment, we consider the mode distinguishability problem for a class of hybrid systems that can describe compromised CPSs. It is found that even though some modes of the hybrid system may not be distinguishable without knowing the attack inputs, the modes could be controlled distinguishable, which means their behaviors can be differentiated under certain control inputs. In this letter, the characterization of controlled distinguishability is studied, and the problem of finding control inputs for mode identification is proposed.

Low latency and high throughput are two of the most critical performance requirements for big data stream computing systems. As multi-source high-speed data streams arrive in real time, it is essential to study latency-aware and resource-aware scheduling to reduce latency and increase throughput. In this paper, we propose a latency- and resource-aware scheduling framework (Lr-Stream) targeting stream-oriented big data applications. Our key contributions can be summarized as follows: (1) a stream topology model and resource scheduling model Lr-Stream are proposed, aiming at optimizing latency and throughput; (2) a latency-aware scheduling strategy and a resource-aware scheduling strategy are proposed; (3) Lr-Stream together with monitor, calculator, and deployment function modules are implemented, and integrated into Apache Storm; (4) system metrics are thoroughly evaluated from latency- and resource-aware perspective on a typical distributed stream computing platform. Experimental results demonstrate that the proposed Lr-Stream yields significant performance improvements in terms of reducing system latency and increasing system throughput.

Data compression is of great significance to the storage of power dispatch data. The compressed data not only saves a lot of space, but also improves the speed of data retrieval. Therefore, this paper proposes an improved algorithm based on run length encoding to compress power dispatch data. In the case of high data similarity, run length coding and dictionary coding have good compression effects, so this article classifies power dispatch data, and uses an improved run length coding algorithm to compress data with high similarity. For similarity Low data is compressed using the Lz4 algorithm that comes with HBase, and a Base-128 Varints encoding the number of occurrences of elements in the run-length encoding is proposed to further improve the compression rate of similar data. This improved algorithm is proven in experiments In some scenarios, the compression rate is indeed improved.

An elastic stream computing system needs elastic adjustment of computing resource allocation and vertex parallelism to improve latency and throughput, which includes continuously or periodically scaling in/out the workload of computing nodes at runtime. Dynamic redirection can help with this elasticity issue by dynamically redirecting real-time data streams to computing resources. Due to the time-varying and unpredictable nature of real-time data streams, implementing redirection of data streams is challenging. Currently, the requirements of data streams redirection are not fully fulfilled, which directly affects the latency and throughput of stream computing systems. To bridge this gap, we proposed a dynamic redirection framework (called Dr-Stream) for elastic stream computing systems. This paper discussed the following aspects: (1) Investigating the dynamic redirection of real-time data streams, providing a general stream application model, a data stream model and a data stream grouping model, as well as formalizing the problem of load balancing optimization and data stream redirection. (2) Redirecting data streams among multiple instances of an operator at runtime by a lightweight load balancing strategy to improve the load balancing of a data center at the vertex level. Managing system states, especially the states of stateful operators by a logical ring-based strategy to improve accuracy. (3) Determining the number of instances for each operator, and deploying the instance(s) to computing nodes by a modified first-fit strategy at runtime. (4) Evaluating the fulfillment of low latency, high throughput, and load balancing objectives in a real-world distributed stream computing environment. Experimental results showed that Dr-Stream reduced the average system latency and load balancing of the data center by more than 20% and 15%, respectively. It also improved the average system stability by more than 15% and avoided over-utilization of computing nodes, as compared to the existing strategies in Storm.

The air traffic management system is one of the most complex manmade systems, with stringent standards for safety and operational performance. Modern surveillance systems make available detailed fl...

Android applications are developing rapidly across the mobile ecosystem, but Android malware is also emerging in an endless stream. Many researchers have studied the problem of Android malware detection and have put forward theories and methods from different perspectives. Existing research suggests that machine learning is an effective and promising way to detect Android malware. Notwithstanding, there exist reviews that have surveyed different issues related to Android malware detection based on machine learning. We believe our work complements the previous reviews by surveying a wider range of aspects of the topic. This paper presents a comprehensive survey of Android malware detection approaches based on machine learning. We briefly introduce some background on Android applications, including the Android system architecture, security mechanisms, and classification of Android malware. Then, taking machine learning as the focus, we analyze and summarize the research status from key perspectives such as sample acquisition, data preprocessing, feature selection, machine learning models, algorithms, and the evaluation of detection effectiveness. Finally, we assess the future prospects for research into Android malware detection based on machine learning. This review will help academics gain a full picture of Android malware detection based on machine learning. It could then serve as a basis for subsequent researchers to start new work and help to guide research in the field more generally.

The burgeoning of Internet of Things (IoT) and camera-equipped mobile devices contributes a tremendous amount of video data generated at the edge of the network. At the same time, we have witnessed the fast deployment of many video-based application services, such as plate recognition for public safety, intelligent transportation, Industry 4.0 and so on. The success of these services, in turn, requires large-scale video data being learned, stored, and retrieved in a more efficient way. A generic software and hardware framework for large-scale IoT video analysis and service support is still missing. To address this challenge, we present π -Hub, PerceptIn’s robotic cloud solution which supports large-scale video data analysis, storage, and query by implementing the learn-store-retrieve paradigm. Interestingly, we found that among the learning, storage, and retrieval services each of them stresses one type of resources on heterogeneous computing servers, i.e., GPU, CPU, and Memory, respectively, therefore it is extremely cost-efficient to co-locate these services together to fully utilize the resources. In addition, several optimization techniques for data writing, reading, and data reduction are proposed and evaluated. The evaluation results show that these techniques improve the performance of the learning, storage and retrieval services significantly as well as notably reduce the cost of the system. We also verify π -Hub’s scalability by reliably running a 1000-machine deployment to support up to one million users. Finally, we conclude the paper by discussing several lessons learned from this study and future work.

In this paper, we propose a human-automation interaction scheme to improve the task performance of novice human users with different skill levels. The proposed scheme includes two interaction modes: learn from experts mode and assist novices mode. In the learn from experts mode, the automation learns from a human expert user such that the awareness of task objective is obtained. Based on the learned task objective, in the assist novices mode, the automation customizes its control parameter to assist a novice human user towards emulating the performance of the expert human user. We experimentally test the proposed human-automation scheme in a designed quadrotor simulation environment, and the results show that the proposed approach is capable of adapting to and assisting the novice human user to achieve the performance that emulates the expert human user.

The automatic extraction of buildings from remote sensing images plays an important role in many areas such as urban planning, geographic information system, and military reconnaissance. This paper presents an automatic building extraction method using the full convolutional neural network and conditional random fields (CRF). Firstly, a deep fully convolutional neural network that fits the building extraction task is designed and trained based on the U-net network structure. Secondly, based on the preliminary results produced by the deep fully convolutional network, the method of conditional random fields is used to optimize the extraction results, which improves the extraction efficiency of large building profiles. The experimental results show that the proposed method can accurately realize the automatic extraction of buildings from remote sensing images, with the recall, precision, and F1-score respectively 75.8%, 85%, and 80.1%. This method has good performance and no artificial design features, so it can be effectively applied to extract building targets from remote sensing images.

This paper investigates the fixed-wing aircraft flocking problem with nonholonomic constraints, speed limits, conflict avoidance, and the efficient use of airspace taken into account. It is challen...

State and runtime-aware scheduling is one of the problems that is hard to resolve in elastic big data stream computing systems, as the state of each vertex is different, and the arrival rate of data streams fluctuates over time. A state and runtime-aware scheduling framework should be able to dynamically adapt to the fluctuation of the arrival rate of data streams and be aware of vertex states and resource availability. Currently, there is an increasing number of research work focusing on application scheduling in stream computing systems, however, this problem is still far from being completely solved. In this paper, we focus on the state of vertex in applications and the runtime feature of resources in a data center, and propose a state and runtime-aware scheduling framework (Sra-Stream) for elastic streaming computing systems, which incorporates the following features: (1) Profiling mathematical relationships between the system response time and the arrival rate of data streams, and identifying relevant resource constraints to meet the low response time and high throughput objectives. (2) Classifying vertex into stateless vertex or stateful vertex from a quantitative perspective, and achieving vertex parallelization by considering the state of the vertex. (3) Demonstrating a proposed stream application scheduling scheme consisting of a modified first-fit based runtime-aware data tuple scheduling strategy at the initial stage, and a maximum latency-sensitive based runtime-aware data stream scheduling strategy at the online stage, by considering the current scheduling status of the application. (4) Evaluating the achievement levels of low response time and high throughput objectives in a real-world elastic stream computing system. Experimental results conclusively demonstrate that the proposed Sra-Stream provides significant performance improvements on achieving the low system response time and high system throughput.

We propose a unified mathematical model for multilayer-multiframe compressive light field displays that supports both attenuation-based and polarization-based architectures. We show that the light field decomposition of such a display can be cast as a bound constrained nonlinear matrix optimization problem. Efficient light field decomposition algorithms are developed using the limited-memory BFGS (L-BFGS) method for automultiscopic displays with high resolution and high image fidelity. In addition, this framework is the first to support multilayer polarization-based compressive light field displays with time multiplexing. This new architecture significantly reduces artifacts compared with attenuation-based multilayer-multiframe displays; thus, it can allow the requirements regarding the number of layers or the refresh rate to be relaxed. We verify the proposed methods by constructing two 3-layer prototypes using high-speed LCDs, one based on the attenuation architecture and one based on the polarization architecture. Moreover, an efficient CUDA-based program is implemented. Our displays can produce images with higher spatial resolution with thinner form factors compared with traditional automultiscopic displays in both simulations and experiments.

Aiming at the dynamic problems caused by the ultra-high speed impact of tiny debris on the large-aperture space deployable antenna, as well as a deeper understanding of the dynamic characteristics of the large-aperture antenna when it works in orbit, the dynamic characteristics of deployable antenna are analyzed. Firstly, ANSYS finite element simulation software is used to establish a spatial deployable antenna structure model to simulate various working environments of the antenna structure in space. Secondly, the dynamic characteristics of the antenna structure are obtained by the modal analysis of the antenna structure using the subspace method. Thirdly, the dynamic response of the antenna structure and the damage state of the key components under the impact of different impact velocities are studied respectively. The results show that the antenna structure has the characteristics of low stiffness, large flexible amplitude, low frequency and small self-damping. In the design of deployable antenna structure, corresponding improvement measures can be given to improve the stiffness and durability of the structure. The dynamic analysis of the space deployable antenna structure provides a theoretical basis for optimizing the dynamic characteristics of the structure and improving the working performance of the structure, and also provides a favorable reference for the design of the large aperture space deployable antenna structure.

In order to study the influence of temperature on the structure of deployable antenna, a dynamic analysis method was proposed. ANSYS finite element software was used to establish the modular space deployable antenna support truss structure model, subspace method was used to conduct modal analysis of the space deployable support truss structure, the natural frequency and mode of the structure was obtained. At the same time, the temperature environment of the deployable antenna in orbit is simulated, the deflection of the structure at 100 temperature is analyzed, and the deflection diagram of the support truss structure at 100 temperature is obtained. The results show that temperature has a great influence on the deflection of the antenna structure, and corresponding measures should be taken to reduce the influence of temperature on the deflection of the antenna structure during the design of deployable antenna. The dynamic analysis method proposed in this paper can provide reference for other truss deployable antenna structures.

Sentiment analysis of film reviews is the basis of obtaining the opinions of movie viewers. It has an important influence on movie public opinion control and stimulating potential viewers. Due to the natural openness and randomness of social media, there may exist a considerable amount of useless or false information in film review comments, making it challenging to analyze the credibility of the comments. This paper proposes a fine-grained sentiment analysis method based on the key-viewpoint sentences of Chinese film reviews, where a deep learning model is used to classify the fine-grained emotions in film reviews. Based on the analysis results, a method for calculating the credibility of review comments is proposed. Under the credibility criteria, corpus screened through credibility filtering algorithm, the overall sentiment classification can obtain 9% improvement on accuracy than the original corpus, which verifies the validity of the credibility algorithm. The higher quality corpus achieved by the credibility algorithm is benefit for improving the accuracy of the sentiment classification.

Electronic documents can carry lots of information and are widely used in daily lives. It will cause substantial economic losses to individual users, enterprises, and governments when the documents containing sensitive information are leaked. How to detect sensitive information to prevent data leakage is still a challenge in the field of information security. This paper mainly focuses on the detection of unstructured documents containing sensitive information. Governments, military, and other institutions can actively mark whether the electronic documents contain sensitive information according to the detection results. We propose a reliable method to detect sensitive electronic documents automatically and compare it with other basic methods. The algorithm structure can extract the characteristics of the data more comprehensively to obtain better detection results. Our model outperformed the other models with 93.44 % accuracy. Our model can also reduce the time cost, which is beneficial for realistic production.

Rapid increase in energy consumption is a serious problem in cloud storage systems. Data accessed in large-scale storage systems usually exhibit temporal and spatial characteristics, which make it possible to reduce energy consumption by clustering data with similar access characteristics for storage in the same zone of cloud storage systems. Existing works usually only focus on the frequency of data access. However, widely existing phenomena show data access with seasonal and tidal characteristics in cloud storage systems. The seasonal and tidal characteristics of data access are extracted thoroughly in this paper. According to the extracted data access characteristics, energy-aware data clustering through a machine learning algorithm (K-ear) is proposed. K-ear classifies data into five seasonal categories according to their seasonal access characteristics and then classifies every seasonal category into three tidal categories according to its tidal access characteristics. The 15 classified categories are stored in different storage zones with different energy and performance modes. Simulation experiments using CloudSimDisk with the constructed mathematic models demonstrate that the proposed K-ear algorithm is more energy-efficient than the default data clustering algorithms in Hadoop and the classical data clustering storage strategy according to the data access frequency (Striping-Based Energy-Aware Strategy).

Knowledge Distillation (KD) based methods adopt the one-way Knowledge Transfer (KT) scheme in which training a lower-capacity student network is guided by a pre-trained high-capacity teacher network. Recently, Deep Mutual Learning (DML) presented a two-way KT strategy, showing that the student network can be also helpful to improve the teacher network. In this paper, we propose Dense Cross-layer Mutual-distillation (DCM), an improved two-way KT method in which the teacher and student networks are trained collaboratively from scratch. To augment knowledge representation learning, well-designed auxiliary classifiers are added to certain hidden layers of both teacher and student networks. To boost KT performance, we introduce dense bidirectional KD operations between the layers appended with classifiers. After training, all auxiliary classifiers are discarded, and thus there are no extra parameters introduced to final models. We test our method on a variety of KT tasks, showing its superiorities over related methods. Code is available at https://github.com/sundw2014/DCM.

In this paper, the game theoretical analysis method is presented to provide optimal strategies for anomaly-based intrusion detection systems (A-IDS). A two-stage game model is established to represent the interactions between the attackers and defenders. In the first stage, the players decide to do actions or keep silence, and in the second stage, attack intensity and detection threshold are considered as two important strategic variables for the attackers and defenders, respectively. The existence, uniqueness, and explicit computation of the Nash equilibrium are analyzed and obtained by considering six different scenarios, from which the optimal detection and attack actions are provided. Numerical examples are provided to validate our theoretical results.

In the mobile crowd sensing network, the existing research does not consider the completion quality factor of the task and the individualized difference of the participant’s ability. The location distribution of the participant will affect the quality of the task and the timeliness of obtaining the sensing task information. Participants in good positions can improve the completion rate of tasks, while participants with good reputation values can ensure the quality of the tasks. In this paper, the distance between the sensing point and the worker is used as one of the criteria for selecting the sensing task object. A utility-based location-distribution reverse auction incentive mechanism (ULDM) is proposed, which comprehensively considers budget constraints, worker’s reputation, and location characteristics in the sensing model, define the distance correlation and time correlation to evaluate the utility of the data collected by the winner. Finally the experimental results show that the successful package delivery rate, average delay and energy consumption are used as evaluation parameters, which improves the quality of task completion and suppresses the selfish behavior of selfish workers, which proves that ULDM has better incentive effect than reputation incentive mechanism.

By simulating the operating dynamics of synchronous generators (SGs), the use of virtual synchronous generators (VSG) can help overcome inverter-based generators' shortcomings of low inertia and minimal damping for grid-forming applications. VSGs' stability is very important for its solar and wind electricity applications. Currently, the related research mostly focuses on VSGs and their application to microgrids. There has been little research to explore how VSGs effect low frequency oscillations in power transmission systems. This paper describes a small-signal model of a VSG-SG interconnected system, which is suitable for studying low frequency oscillations damping in a power transmission grid. Based on the model, the effects of VSGs on low frequency oscillations are compared with the effects of SG to reveal the mechanism of how VSGs influence the damping characteristic. The influence of each VSG control loop on oscillations is also analyzed in the paper. Then, the low frequency oscillation risks with different types of VSGs are described. Finally, experiments on a real-time laboratory (RT-LAB) platform were conducted to verify the small-signal analysis results.

Due to the rapid development of technologies for small unmanned aircraft systems (sUAS’s), the supply and demand market for sUAS’s is expanding globally. With the great number of sUAS’s ready to fly in civilian airspace, an sUAS aircraft traffic management system that can guarantee the safe and efficient operation of sUAS’s is still absent. In this paper, we propose a control protocol design and analysis method for sUAS traffic management (UTM) which can safely manage a large number of sUAS’s. The benefits of our approach are two-fold: at management level, the effort for monitoring sUAS traffic (authorities) and control/planning for each sUAS (operator/pilot) are both greatly reduced under our framework; and at operational level, the behavior of individual sUAS is guaranteed to follow the restrictions. Mathematical proofs and numerical simulations are presented to demonstrate the proposed method.

A hybrid systems approach is a powerful tool for modeling and analysis of cyber-physical systems (CPSs), but it has been rarely used for CPS security related problems. In this paper, we model the CPS subject to cyber-attacks as a class of hybrid systems called hidden mode switched linear systems so that it can account for both the switching attack that tampers the discrete state dynamics (logical behavior) and the bounded data injection attack which compromises the continuous state dynamics (physical behavior) of the CPS. A resilient hybrid control scheme is proposed to mitigate the impact of these cyber-attacks. It is shown that the proposed resilient hybrid control scheme is able to estimate the switching attack signal and guarantee the attack mitigation under the assumptions that the discrete states of the hybrid system are distinguishable and the switching attack signal satisfies the average dwell-time condition.

The numerical flux is an important element in the numerical model for solving shallow water flow problems in MUSCL scheme. It is used to determine the normal transport through the boundary of the control body. With the TVD numerical flux, because of the simple form of Lax–Friedrichs, it is adopted in many articles. In fact, there are many different numerical fluxes based on exact Riemann solutions or approximate Riemann solutions, and these numerical fluxes can be combined with the MUSCL scheme. For the one-dimensional and two-dimensional shallow water equations, starting from the accuracy and discontinuous capture ability, a series of numerical examples are simulated based on the six different numerical fluxes with MUSCL scheme. Through the result analysis, the paper find out the numerical fluxes which are more suitable for shallow water flow combined with MUSCL scheme.

Most of the numerical methods used for shallow water flow calculation are not applicable to the numerical calculation of the discontinuous flows. High precision resolution is a difficult problem in simulate discontinuous flows. This paper adopts Roe scheme approximate Riemann solver to deal with normal vector through interface, apply TVD–MUSCL scheme and Hancock scheme to increase the precision of space and time to second order using in tidal bore simulation. In order to verify the accuracy of the model and simulate the ability of discontinuous flows, some numerical examples are simulated in this paper, the results of simulation and other scholars’ results are compared to check the applicability and accuracy of the model for discontinuous flow. With the generalized river course tidal bore experiment to prove the model has the ability to capturing shock wave and better calculative precision.

This article describes how with the continuous expansion on the volume of data produced by sensors in Cyber Physical Systems, the scale of the cloud storage system has become larger. This will lead to the problems of a high energy consumption rate and a low utilization becoming a serious issue. In order to enhance the effective energy consumption, reduce the invalid energy consumption, and supply more flexible QoS for users in CPS, this article proposes an automatic energy gear-shifting mechanism with flexible QoS constraints (QGLG). The QGLG predicts system load of the follow-up period through a support vector machine model. According to the current system load, the predicted load, and the flexible QoS, QGLG automatically up-shifts and down-shifts among nodes. Substantive results from the simulation experiments done on GridSim show that the QGLG can achieve energy consumption reduction while satisfying the user's flexible QoS requirements. Compared with a similar energy-reducing mechanism, QGLG has its obvious advantage when considering the requirements of user with energy saved notwithstanding.

Big data stream computing systems should work continuously to process streams of on-line data. Therefore, fault tolerance is one of the key metrics of quality of service in big data stream computing. In this paper, we propose a fault tolerant framework with deadline guarantee for stream computing called FTDG. First, FTDG identifies the critical path of a data stream graph at a given data stream throughput, and quantifies the system reliability of a data stream graph. Second, FTDG allocates tasks by the fault tolerance aware heuristic and critical path scheduling mechanism. Third, FTDG online optimizes the task scheduling by reallocating the critical vertices on the critical path of the data stream graph to lower the response time and reduce system fluctuations. Theoretical as well as experimental results demonstrate that the FTDG makes a desirable trade-off between high fault tolerance and low response time objectives in big data stream computing environments.

Online scheduling plays a key role for big data streaming applications in a big data stream computing environment, as the arrival rate of high-velocity continuous data stream might fluctuate over time. In this paper, an elastic online scheduling framework for big data streaming applications (E-Stream) is proposed, exhibiting the following features. (1) Profile mathematical relationships between system response time, multiple application fairness, and online features of high-velocity continuous stream. (2) Scale out or scale in a data stream graph by quantifying computation and communication cost, and the vertex semantics for arrival rate of data stream, and adjust the degree of parallelism of vertices in the graph. Subgraph is further constructed to minimize data dependencies among the subgraphs. (3) Elastically schedule a graph by a priority-based earliest finish time first online scheduling strategy, and schedule multiple graphs by a max–min fairness strategy. (4) Evaluate the low system response time and acceptable applications fairness objectives in a real-world big data stream computing environment. Experimental results conclusively demonstrate that the proposed E-Stream provides better system response time and applications fairness compared to the existing Storm framework.

Deep learning can enable Internet of Things (IoT) devices to interpret unstructured multimedia data and intelligently react to both user and environmental events but has demanding performance and power requirements. The authors explore two ways to successfully integrate deep learning with low-power IoT products.

Mobile payment becomes more and more popular, and the main way of mobile payment, third-party payment, still faces many security problems. Based on the mainstream Android platform, this paper studies the payment process of the third-party payment platform and analyzes the security defects and security problems that may occur when the application developers realize the third-party payment process. The paper tested the security defects through experiments, and the test results showed that the integration of third-party payment functions was relatively popular in applications, and many applications integrating third-party payment functions had different degrees of security defects.

Establishing correspondences between two images requires both local and global spatial context. Given putative correspondences of feature points in two views, in this paper, we propose Order-Aware Network, which infers the probabilities of correspondences being inliers and regresses the relative pose encoded by the essential matrix. Specifically, this proposed network is built hierarchically and comprises three novel operations. First, to capture the local context of sparse correspondences, the network clusters unordered input correspondences by learning a soft assignment matrix. These clusters are in a canonical order and invariant to input permutations. Next, the clusters are spatially correlated to form the global context of correspondences. After that, the context-encoded clusters are recovered back to the original size through a proposed upsampling operator. We intensively experiment on both outdoor and indoor datasets. The accuracy of the two-view geometry and correspondences are significantly improved over the state-of-the-arts. Code will be available at https://github.com/zjhthu/OANet.git., Comment: Accepted to ICCV 2019, and Winner solution to both tracks of CVPR IMW 2019 Challenge. Code will be available soon at https://github.com/zjhthu/OANet.git

The issue of highly efficient geographically-distributed autonomous data management is one of the critical obstacles for opening up a big data era. It creates the need for investigating autonomous and distributed data management systems in big data environments. In this paper, a distributed autonomous data management system is put forward, exhibiting the following features. (1) A distributed architecture designed to meet the requirements of autonomous data management by allowing interconnection, intercommunication, and interoperation of multiple sites over the Internet. (2) An autonomous, multi-level, unstructured data storage system to meet high-efficiency storage needs, with reference to the distributed heterogeneous data storage theories. (3) A distributed autonomous data indexing and retrieval system to support metadata search & full-text searching, fast loading, remote access, and unified view. Experimental and application results demonstrate that the proposed system has high potential to reduce access time and improve storage efficiency, while maintaining satisfactory availability and scalability.

With the massive growth of big data applications, the requirement for data processing speed is getting higher and higher in stream computing systems. The Storm, as one of the most popular distributed stream computing systems, has received more attention. However, the Storm's traditional scheduling strategy is not ideal for processing a large volume of streaming data. The resource scheduling in a distributed stream computing system should consider not only node allocation status but also fluctuating input rates of data stream. To address this problem, this paper has completed the following work: (1) A performance model La-Stream (latency-sensitive elastic adaptive scheduling) is proposed and built by adopting a quantitative method for calculating the amount of computation required between task map nodes and node communication. (2) A La-Stream based algorithm is proposed. The algorithm dynamically plans a resource allocation scheme with minimal data processing latency among available resources to achieve optimal allocation. (3) Three functional modules of La-steam are proposed and implemented: module Monitor, module Optimizer and module Scheduler. The three modules are integrated into the Storm platform with minimal overhead. Several sets of experiments are conducted, verifying the feasibility and effectiveness of La-Stream.

The access of wind turbines to the power grid will bring new challenges to the safe and stable operation of power systems. When the wind turbine adopts the virtual synchronous control strategy, it can match the characteristics of the traditional synchronous generator and solve the unstable operation problem caused by the lack of inertia. In this paper, the influence mechanism of virtual synchronous machine (VSM) model on voltage stability of power system is analyzed with bifurcation theory. The increase of the damping coefficient of the VSM and the reduction of the inertia time constant will be beneficial to the voltage stability, which can provide a basis for the wind power transmission limit.

Convolutional Neural Networks (CNNs) have become deeper and more complicated compared with the pioneering AlexNet. However, current prevailing training scheme follows the previous way of adding supervision to the last layer of the network only and propagating error information up layer-by-layer. In this paper, we propose Deeply-supervised Knowledge Synergy (DKS), a new method aiming to train CNNs with improved generalization ability for image classification tasks without introducing extra computational cost during inference. Inspired by the deeply-supervised learning scheme, we first append auxiliary supervision branches on top of certain intermediate network layers. While properly using auxiliary supervision can improve model accuracy to some degree, we go one step further to explore the possibility of utilizing the probabilistic knowledge dynamically learnt by the classifiers connected to the backbone network as a new regularization to improve the training. A novel synergy loss, which considers pairwise knowledge matching among all supervision branches, is presented. Intriguingly, it enables dense pairwise knowledge matching operations in both top-down and bottom-up directions at each training iteration, resembling a dynamic synergy process for the same task. We evaluate DKS on image classification datasets using state-of-the-art CNN architectures, and show that the models trained with it are consistently better than the corresponding counterparts. For instance, on the ImageNet classification benchmark, our ResNet-152 model outperforms the baseline model with a 1.47% margin in Top-1 accuracy. Code is available at https://github.com/sundw2014/DKS., Comment: Added supplementary materials, and the code is available at https://github.com/sundw2014/DKS

With the increasing popularity of location technologies and location-based service applications, a large number of user's trajectory data have been collected. Publishing the real-time statistics data of trajectory streams can be useful in many fields such as intelligent transportation system, urban road planning and road congestion detection. As the trajectory data itself contains a wealth of user's privacy information, the privacy leakage problem has aggravated the risk of data publishing. In order to realize the personalized and uniform privacy preserving of user's trajectory data, the differential privacy model based on data perturbation is introduced, and a privacy preserving algorithm is proposed. The algorithm contains three modules of dynamic privacy budget allocation, privacy approximation and privacy publishing. In the experiment, the performances of the proposed method are verified by using real-life datasets.

Automatic target recognition in infrared imagery is a challenging problem. In this paper, a kernel sparse coding method for infrared target recognition using covariance descriptor is proposed. First, covariance descriptor combining gray intensity and gradient information of the infrared target is extracted as a feature representation. Then, due to the reason that covariance descriptor lies in non-Euclidean manifold, kernel sparse coding theory is used to solve this problem. We verify the efficacy of the proposed algorithm in terms of the confusion matrices on the real images consisting of seven categories of infrared vehicle targets.

The issue of high system stability is one of the major obstacles for real-time computing over fluctuating big data streams. A stable scheduling is more important than an efficient scheduling for stream applications, especially when a scheduling is to be rescheduled dynamically at runtime. In this paper, a stable online scheduling strategy with makespan guarantee SOMG is discussed, which includes the following features: 1) profiling mathematical relationships between system stability, response time, and resource utilization, and indicating conditions to meet the high system stability and acceptable response time objectives; 2) optimizing the structure of a data stream graph by quantifying and adjusting vertices of the graph; and 3) scheduling a data stream graph with heuristic critical path scheduling mechanism, which is subject to response time constraints, rescheduling only key vertices on dynamically changing critical path of the graph, and considering the historical information of current scheduling to maximize system stability with response time aware. Experimental results conclusively demonstrate that the SOMG framework has higher potential of providing enhancement on efficient system stability and guaranteeing significant response time. It efficiently and effectively makes a tradeoff between high system stability and acceptable response time objectives in big data stream computing environments.

One of the issues for tour planning applications is to adaptively provide personalized advices for different types of tourists and tour activities. This paper proposes a high level Petri Nets based approach to providing some level of adaptation by implementing adaptive navigation in a tour node space. The new model supports dynamic reordering or removal of tour nodes along a tour path; it supports multiple travel modes and incorporates multimodality within its tour planning logic to derive adaptive tour. Examples are given to demonstrate how to realize adaptive interfaces and personalization. Future directions are also discussed at the end of this paper.

Performance-aware deployment of streaming applications is one of the key challenging problems in distributed stream computing systems. We proposed a performance-aware deployment framework (Pa-Stream) for distributed stream computing systems. By addressing the important aspects of the framework, this paper makes the following contributions: 1) investigated the performance-aware deployment of a streaming application over distributed and heterogeneous computing nodes, and provided a general application deployment model; 2) demonstrated a streaming applications deployment scheme by proposing an artificial bee colony strategy that deploys application's vertices onto the best set of computing nodes; an incremental online redeployment strategy was used to redeploy the running application; 3) developed and integrated Pa-Stream into Apache Storm platform; 4) evaluated the fulfilment of low latency and high throughput objectives in a distributed stream computing environment. Experimental results demonstrate that the proposed Pa-Stream provided effective performance improvements on latency, throughput and resource utilisation.

In view of the negative influence of selfish nodes in Mobile Crowd Sensing, this paper proposes an incentive mechanism based on Reputation and Trust (RTM). Firstly, this paper analyzes the reputation incentive mechanism and trust incentive mechanism. Secondly, this paper constructs an incentive model, which is divided into user selection module and reward implementation module, and defines the service quality factor, link reliability factor and time heat factor as the pricing factors to calculate comprehensive pricing to reward service providers. Finally, the experimental results show that with successful package delivery rate, average delay and energy consumption as evaluation parameters, in terms of motivating nodes to participate in network cooperation and suppressing the selfish behavior of selfish nodes, it is proved that RTM has better effect and feasibility than reputation incentive mechanism and trust incentive mechanism.

In big data era, fresh data grows rapidly every day. More than 30,000 gigabytes of data are created every second and the rate is accelerating. Many organizations rely heavily on real time streaming, while big data stream computing helps them spot opportunities and risks from real time big data. Storm, one of the most common online stream computing platforms, has been used for big data stream computing, with response time ranging from milliseconds to sub-seconds. The performance of Storm plays a crucial role in different application scenarios, however, few studies were conducted to evaluate the performance of Storm. In this paper, we investigate the performance of Storm under different application scenarios. Our experimental results show that throughput and latency of Storm are greatly affected by the number of instances of each vertex in task topology, and the number of available resources in data center. The fault-tolerant mechanism of Storm works well in most big data stream computing environments. As a result, it is suggested that a dynamic topology, an elastic scheduling framework, and a memory based fault-tolerant mechanism are necessary for providing high throughput and low latency services on Storm platform.

With the development and application of next-generation sequencing (NGS) and target capture technology, the demand for an effective analysis method to accurately detect gene fusion from high-throughput data is growing. Hence, we developed a novel fusion gene analyzing method called single-end gene fusion (SEGF) by starting with single-end DNA-seq data. This approach takes raw sequencing data as input, and integrates the commonly used alignment approach basic local alignment search tool (BLAST) and short oligonucleotide analysis package (SOAP) with stringent passing filters to achieve successful fusion gene detection. To evaluate SEGF, we compared it with four other fusion gene discovery analysis methods by analyzing sequencing results of 23 standard DNA samples and DNA extracted from 286 lung cancer formalin fixed paraffin embedded (FFPE) samples. The results generated by SEGF indicated that it not only detected the fusion genes from standard samples and clinical samples, but also had the highest accuracy and sensitivity among the five compared methods. In addition, SEGF was capable of detecting complex gene fusion types from single-end NGS sequencing data compared with other methods. By using SEGF to acquire gene fusion information at DNA level, more useful information can be retrieved from the DNA panel or other DNA sequencing methods without generating RNA sequencing information to benefit clinical diagnosis or medication instruction. It was a timely and cost-effective measure with regard to research or diagnosis. Considering all the above, SEGF is a straightforward method without manipulating complicated arguments, providing a useful approach for the precise detection of gene fusion variation.

As an important distributed real-time computation system, Storm has been widely used in a number of applications such as online machine learning, continuous computation, distributed RPC, and more. Storm is designed to process massive data streams in real time. However, there have been few studies conducted to evaluate the performance characteristics clusters in Storm. In this paper, we analyze the performance of a Storm cluster mainly from two aspects, hardware configuration and parallelism setting. Key factors that affect the throughput and latency of the Storm cluster are identified, and the performance of Storm’s fault-tolerant mechanism is evaluated, which help users use the computation system more efficiently.

The issue of shortest path discovery in consideration of obstacle is one of the problems for location-based services in mobile social network environments. Currently, most research focuses on quickly discovering the shortest path in obstacle free area with reasonable latency, while the obstacle issue, especially the obstacles that enter temporarily is not fully considered. This creates the need for investigation on shortest path discovery at the same time avoiding detected obstacles. In this paper, a shortest path discovery approach is proposed. The following contributions are made: (1) Modeling the shortest path discovery problem in consideration of obstacle. (2) Discovering the shortest path using an improved A-star algorithm with reasonable latency. (3) Evaluating the accuracy rate of shortest path discovery with acceptable latency for a location-based service in a mobile social network. Experimental results conclusively demonstrate the efficiency and effectiveness of the proposed approach.

To achieve high energy efficiency and low response time in big data stream computing environments, it is required to model an energy-efficient resource scheduling and optimization framework. In this paper, we propose a real-time and energy-efficient resource scheduling and optimization framework, termed the Re-Stream. Firstly, the Re-Stream profiles a mathematical relationship among energy consumption, response time, and resource utilization, and obtains the conditions to meet high energy efficiency and low response time. Secondly, a data stream graph is modeled by using the distributed stream computing theories, which identifies the critical path within the data stream graph. Such a methodology aids in calculating the energy consumption of a resource allocation scheme for a data stream graph at a given data stream speed. Thirdly, the Re-Stream allocates tasks by utilizing an energy-efficient heuristic and a critical path scheduling mechanism subject to the architectural requirements. This is done to optimize the scheduling mechanism online by reallocating the critical vertices on the critical path of a data stream graph to minimize the response time and system fluctuations. Moreover, the Re-Stream consolidates the non-critical vertices on the non-critical path so as to improve energy efficiency. We evaluate the Re-Stream to measure energy efficiency and response time for big data stream computing environments. The experimental results demonstrate that the Re-Stream has the ability to improve energy efficiency of a big data stream computing system, and to reduce average response time. The Re-Stream provides an elegant trade-off between increased energy efficiency and decreased response time effectively within big data stream computing environments.

Trend and outlier are frequently used to derive early warning predictive signal to decision maker in order to achieve ultimate quality decision outcome in domain specific (e.g. commercial, scientific, biomedical and engineering, just to name a few) applications. We develop a gradient-based algorithm using sample entropy gradient(SEG) for trend and outlier prediction in high frequency time series data streams. L2 similarity measure (Euclidean distance between two linearized gradient curves is then computed and used to quantify the degree of similarity and compared with a threshold L2 value to judge the extend of dissimilarity that would be classified as outlier. SEG algorithm which circumvents the need to pre-specify tolerance parameter in those cross sample entropy (CSE)-based algorithms that invariably involve real domain expert to set the tolerance threshold. We conduct real data experiments on SEG algorithm to two application areas: dynamic wind speed data stream; and financial time series data. Our experiments demonstrated that SEG algorithm can be feasibly used in online implementation to derive predictive early warning signals to domain-specific decision maker.

Sparse representation has been widely researched for image-based classification. However, sparse representation classification directly treats training samples as a dictionary, so it needs a large training set and is time consuming, especially for a large training set. To derive a small dictionary, many dictionary learning algorithms are researched. Thus, object recognition problem is transformed to optimize the sparse representation errors on the compact dictionary. The sparse representation optimization is constraint by l 0 -norm, which is NP-hard problem. Though we can use l 1 -norm minimization instead to work effectively, it is still time consuming for optimization. To make the algorithm discriminative and simultaneously decrease the computational burden, we proposed a fast discriminative collaborative representation–based classification algorithm. The new algorithm incorporated the within-class scatter and the linear classification error terms into the objective function to derive a more discriminative dictionary and simultaneously added collaborative representation mechanism to cut off the time consuming. At the end of this article, we designed two experiments to validate our method using near-infrared and AR visible databases for multimodal face recognition. The results showed that our algorithm outperformance sparse representation–based, collaborative representation–based, and discriminative-KSVD classification algorithms.