Skip to main content

Über dieses Buch

This book constitutes the refereed proceedings of the 8th International Symposium on Parallel Architecture, Algorithm and Programming, PAAP 2017, held in Haikou, China, in June 2017.

The 50 revised full papers and 7 revised short papers presented were carefully reviewed and selected from 192 submissions. The papers deal with research results and development activities in all aspects of parallel architectures, algorithms and programming techniques.



Algorithms and Programming


Ford Motor Side-View Recognition System Based on Wavelet Entropy and Back Propagation Neural Network and Levenberg-Marquardt Algorithm

(Aim) Automatic identification of the car manufacturer in the side-view position can be used for the intelligent traffic monitoring system. Currently, the side-view car recognition did not attract too much attention. (Method) We proposed a novel Ford Motor recognition system. We first captured the car image from the side view. Second, we used wavelet entropy to extract texture features. Third, we employed a back propagation neural network (BPNN) as the classifier. Finally, we employed the Levenberg-Marquardt algorithm to train the classifier. In the experiment, we utilized the 3 × 3-fold cross validation. (Result) This method achieved an overall accuracy of 80% in detecting Ford motors. (Conclusion) This method can detect Ford Motors from the side view effectively. In the future, it may also be used to detect cars of other brands.

Wen-Juan Jia, Shuihua Wang, Huimin Lu, Ying Shao, Elizabeth Lee, Yu-Dong Zhang

Intrusion Detection Based on Self-adaptive Differential Evolution Extreme Learning Machine with Gaussian Kernel

In our everyday life, intrusion detection system(IDS) becomes a promising area of research in the domain of security. With the rapid development of network-based services, IDS can detect the intruders who are not authorized to the present computer system, so IDS has emerged as an essential component and an important technique for network security.In order to conquer the disadvantage of the traditional algorithm for single-hidden layer feedforward neural network (SLFN), an improved algorithm, called extreme learning machine (ELM), is proposed by Huang et al. However, ELM is sensitive to the neuron number in hidden layer and its selection is a difficult-to-solve problem. ELM is an interested area of research for detecting possible intrusions and attacks. In this paper, we propose an improved learning algorithm named self-adaptive differential evolution extreme learning machine with Gaussian Kernel (SaDE-KELM) for classifying and detecting the intrusions. We compare our methods with commonly used ELM, DE-ELM techniques in classifications. Simulation results show that the proposed SaDE-KELM approach achieves higher detection accuracy in classification case.

Junhua Ku, Bing Zheng

Prediction for Passenger Flow at the Airport Based on Different Models

Correctly predicting the passenger flow of an air route is crucial for the construction and development of an airport. Based on the passenger flow data of Sanya Airport from 2008 to 2016, ARMA Model, Grey Prediction GM (1, 1) Model and ARMA-improved Regression Model were adopted for data fitting. Upon verification, the average absolute percentage error of such three models was 4.19%, 4.20% and 1.97% respectively with high prediction precision. As a result, the passenger flow at Sanya Airport is predicted to reach 20 million within two years.

Xia Liu, Xia Huang, Lei Chen, Zhao Qiu, Ming-rui Chen

Election Based Pose Estimation of Moving Objects

In this work, a key-points based method is presented to track and estimate the pose of rigid objects, which is achieved by using the tracked points of the object to calculate the attitude changes [1]. We propose to select a few points to represent the posture of the object and maintain efficiency. A standard feature point tracking algorithm is applied to detect and match feature points. The presented method is able to overcome key-points’ errors as well as decrease the computational complexity. In order to reduce the error caused by feature points detection, we use the tacked key-points and their relation with the target center to get the most reliable tracking result. To avoid introducing errors, the model will maintain the features generated in initialization. Finally, the most reliable candidates will be picked out to calculate the pose information, and the small amount of key-points with highly accuracy can ensure real-time performance.

Liming Gao, Chongwen Wang

A Novel Topology Reconfiguration Backtracking Algorithm for 2D REmesh Networks-on-Chip

This paper presents a fault-tolerant topology reconfiguration backtracking algorithm to tolerate faulty cores in 2D REmesh based (reconfigurable mesh based) Networks-on-Chip. This new algorithm can be dynamically reconfigured to support irregular topologies caused by faulty cores in a REmesh network without destroying the integrity of topologies. In addition, the proposed reconfigure method has a high-level fault-tolerance capability and therefore it is capable to tolerate more faulty components in more complicated faulty situations without additional hardware costs. The reliability performance and fault-tolerance capability of the reconfiguration backtracking algorithm in a 2D REmesh network are evaluated through appropriate simulations. The experimental results show that in different sizes of topologies (the max size is 7 × 8), when less than 10.7% faulty cores occur, more than 91.5% successful reconfiguration rate can be achieved. In addition, in the 7 × 8 REmesh, when the faulty core reaches 7, the successful reconfiguration rate has reached 61.49%, which enhanced 9.74% compared with the TRARE algorithm.

Na Niu, Fang-Fa Fu, Hang Li, Feng-Chang Lai, Jin-Xiang Wang

User Behaviour Authentication Model Based on Stochastic Petri Net in Cloud Environment

Cloud Computing has been developing rapidly and has impacted various aspects of our daily life. However, the growth of cloud service raises concerns about its security. This paper will be focussing on user identity authentication issues in the cloud environment. Through analysing and classifying user behaviours, we propose a Stochastic Petri net-based User Behaviour Authentication model (SPUBA), which uses the behaviour of a user while logging in and browsing to analyse user behaviour credibility. Also, in order to quantify user behaviour credibility, we have modified a K-modes algorithm to solve user habitual behavioural standard, proposed an algorithm for calculating user behaviour credibility. The user operational behaviours simulations has been performed in the cloud environment to analyse the execution time of the proposed model. The results regarding the detection and false positive rates shown are better than current models.

Peng Li, Cheng Yang, He Xu, Ting Fung LAU, Ruchuan Wang

Performance Prediction of Spark Based on the Multiple Linear Regression Analysis

It is crucial to evaluate performance of a cloud platform and determine the main factors influencing the property. Moreover, the analysis results of related performance indicators can be applied to making theoretical predictions about the performance status of the cloud platform. This work mainly focuses on researching the interrelations between the performance indicators based on the Spark technology of the cloud platform and the load performance of the cluster, and furthermore makes effective predictions for the load performance. Firstly, we put forward the analytic frameworks of Spark performance analysis, the specific indicators analysis as well as the prediction models towards the cluster load. Secondly, with respect to the evaluation indicators, we explore the basis for their selections as well as their concrete implications, and then objectively, accurately calculate the correlation formula between the practically produced performance parameters and the load performance of the cluster when the Spark cluster performs the batch applications utilizing the MLR (Multiple Linear Regression) method, and, therefore, determine the main factors impacting the load performance. Finally, we predict the load value utilizing the Spark indicator analysis and the load prediction model. The results indicate that accuracy is up to 92.307%. Consequently, the solution presented in this paper predicts the cluster load value with effetioncy.

Lu Dong, Peng Li, He Xu, Baozhou Luo, Yu Mi

CTS-SOS: Cloud Task Scheduling Based on the Symbiotic Organisms Search

Cloud task scheduling affects the overall operating efficiency of the cloud platform. Thus, how to effectively use resources in the cloud environment and make massive tasks to implement a reasonable and efficient scheduling becomes more crucial. Firstly, the mathematical model of cloud task computing was reconstructed by adding the expected completion time to the task. Secondly, on the basis of the completion time as the fitness function, the task priority was dynamically adjusted by user satisfaction, which was added to reduce the user’s completion time and improve the user’s satisfaction. Thirdly, aiming at the continuous search space, a cloud task scheduling algorithm based on the Symbiotic Organisms Search (CTS-SOS) was proposed. Not only does the CTS-SOS have fewer specific parameters, but also take a little time complexity. Through using the CloudSim toolkit package, the CTS-SOS algorithm was compared with Round Robin algorithm of the CloudSim and ACO algorithm. Experimental results show that CTS-SOS can provide a better optimization and scheduling of resources, reduce the makespan effectively, and improve the efficiency of processing tasks and user’s satisfaction.

Zhenpeng Liu, Xiaodan Liu, Yawei Dong, Xuan Zhao, Bin Zhang

Exploration of Heuristic-Based Feature Selection on Classification Problems

We present two heuristics for feature selection based on entropy and mutual information criteria, respectively. The mutual-information-based selection algorithm exploiting its submodularity retrieves near-optimal solutions guaranteed by a theoretical lower bound. We demonstrate that these heuristic-based methods can reduce the dimensionality of classification problems by filtering out half of its features in the meantime still improving classification accuracy. Experimental results also show that the mutual-information-based heuristic will most likely collaborate well with classifiers when selecting about a half size of features, while the entropy-based heuristic will help most in the early stage of selection when choosing a relatively small percentage of features. We also demonstrate a remarkable case of feature selection being used in classification on a medical dataset, where it can potentially save half of the cost on the diabetes diagnosis.

Qi Qi, Ni Li, Weimin Li

AGSA: Anti-similarity Group Shilling Attacks

With the rapid development of e-commerce, the security issues of recommender systems have been widely investigated. Malicious users can benefit from injecting great quantities of fake profiles into recommender systems to reduce the frequency of undesired recommendation items. As one of the most important attack methods in recommender systems, the shilling attack has been paid considerable attention, especially to its model and the way to detect it. Although a multitude of studies have been devoted to shilling attack modeling and detection, few of them focus on group shilling attack. The attackers in a shilling group work together to manipulate the output of the recommender system. Based on the model of the loose version of Group Shilling Attack Generation Algorithm (GSAGenl), we design an anti-similarity group shilling attack model (AGSA). AGSA rationalizes the evaluation time interval of the group attack and strengthens the destructive powers of the group shilling attacks.

Peng Wang, Lingtao Qi, Haiping Huang, Feng Li, Congxiang Yu

H ∞ Filtering Design for a Class of Distributed Parameter Systems with Randomly Occurring Sensor Faults and Markovian Channel Switching

The paper is concerned with H filtering design for a class of stochastic distributed parameter systems with randomly occurring sensor faults over sensor networks with multiple communications. The channel switching is governed by a continuous-time Markovian process and the case of measurement failures is described by a stochastic variable which is satisfying the Bernoulli random distribution. Based on a Markovian switched Lyapunov-Krasovskii functional, delay-dependent conditions are achieved to guarantee the prescribed H performance. Finally, a practical simulation example is given to illustrate the validity of our results.

Huihui Ji, Baotong Cui

The Study of the Seabed Side-Scan Acoustic Images Recognition Using BP Neural Network

In recent years, mankind has made great achievements in the marine exploration. Ocean contains abundant resources, and the seabed has recorded amount of basic Earth information. Therefore, a complete study of the seabed can help to form a full appreciation of underwater environment. The study of the seabed recognition method, as the most basic work of the study of the seabed, is gradually gaining the attention of researchers. As a main marine exploratory tool, the side-scan sonar is fast, accurate and convenient for seabed information collection. In this paper, lots of seabed acoustic images were applied to extract the seabed substrate characteristics using the gray covariance matrix method. An improved BP neural network model was involved into classify and identify the seabed characteristics. In addition, several algorithms for BP neural network were proposed for testing the recognition accuracy of side-scan acoustic images and the convergence rate. The results show that although several algorithms were easy to fall into the minimum value during training, which can lead to slow convergence rate and unable to meet the recognition accuracy standard, the trainlm function had a faster convergence rate and higher recognition accuracy.

Hongyan Xi, Lei Wan, Mingwei Sheng, Yueming Li, Tao Liu

Node Localization of Wireless Sensor Network Based on Secondary Correction Error

Due to the large localization error of the range-free localization algorithm, a new node localization algorithm based on secondary correction error is proposed. Firstly, orthogonal polynomial fitting method, a mathematical model, is taken advantage of to correct the distance error. Moreover, subtraction first and then square, a strategy, is introduced to solve the equations. At the same time, the actual distance and distance error are taken as weighting factors to construct the weighted matrix to solve the unknown node coordinates. Finally, the redundant information obtained by solving the equations is employed to refine the coordinates of unknown nodes. Simulation experiment results in this paper are convincing evidence that our proposed algorithm can decrease positioning error and increase the positioning accuracy efficaciously.

Xiaoxu Ma, Wenju Liu, Ze Wang

Optimizations of the Whole Function Vectorization Based on SIMD Characteristics

Vectorization for SIMD extensions is similar to programming for CUDA/OpenCL on GPU platforms. They are both Single Program Multiple Data (SPMD) programming models. However, SIMD extensions and GPU accelerators are different from each other in many aspects, such as memory access, divergence, etc. There are still optimization opportunities when using existing methods to implement vectorization for SIMD extensions. As a result, we propose a whole function vectorization optimization algorithm based on SIMD characteristics in this paper. First, we analyze some SIMD characteristics that may affect the whole function vectorization. These characteristics include instance versioning, instance regrouping and SIMD code optimization. We then implement a SIMD characteristics-based algorithm for whole function vectorization. In addition, we introduce a directive based method to help us fully exploit opportunities of this kind of vectorization. We choose nine benchmarks from multi-media and image processing applications to evaluate our technique. Compared with un-optimized codes, the speedup is 1.59 times faster in average on processor E5-2600 when the proposed technique is applied.

Yingying Li, Yuchen Gao, Dong Wang, Yanbing Li, Jinlong Xu

A Stacked Denoising Autoencoders Based Collaborative Approach for Recommender System

This paper uses an autoencoder neural network as user feature learning component for collaborative filtering task. We propose a stacked denoising autoencoder (SDAE) based model to alleviate the sparseness issues in recommendation system. Our model also extends the scalability of CF-based methods in the Top-N recommendation task. Experiments on MovieLens datasets and the result confirmed the effectiveness and potential of our model.

Baojun Niu, Dongsheng Zou, Yafeng Niu

Research on Adaptive Canny Algorithm Based on Dual-Domain Filtering

In order to overcome the shortcomings of the traditional Canny algorithm, which is easy to lose the edge details of the edge detection, and is prone to a large number of false edges and the threshold needs to be manually determined, an adaptive Canny algorithm based on Dual-domain filtering is proposed. Firstly, the Dual-domain filter is used to instead of the traditional algorithm, which is used to remove the image noise while preserving the image edge information, and more effective than the traditional Canny algorithm; Secondly, the Otsu algorithm is used to calculate the high and low threshold of Canny operator in order to achieve the purpose of high automation. The experimental results show that the improved algorithm can detect more effective edges of the image, but also has a strong adaptability.

Xiajiong Shen, Xiaoyu Duan, Daojun Han, Wanli Yuan

A Dynamic Individual Recommendation Method Based on Reinforcement Learning

As a widely used recommendation method, collaborative filtering can solve the problem of low level of resource utilization which caused by information overload. At present, in order to exhibiting and searching items, we need to use multipole attributes to describe items. Thus request to particularly distinguish every attribute and realize accurate recommendation. While the collaborative filtering method lose sight of the dynamic regulation of items attributes’ importance degree, and it cannot interpose the discrimination of attributes. Aiming at this problem, this paper come up with a dynamic individual recommendation method based on reinforcement learning. This method can dig user’s attribute tag preference from operant behavior. It can record user’s attributes operate path and recall path. Then we build the award-punishment model of attribute tag, and realize the tag weight dynamic regulation. According to the principle that reinforcement learning system always get max award, we make a tag recommend strategy and give user recommendations in accordance with the preferences. The experimental result show that this method can distinguish the validity of user’s click, and realize the tag weight dynamic regulation and give user recommendations in accordance with the preferences.

Daojun Han, Xiajiong Shen, Tian Gan, Ruiqing Cai

Research on the Pre-distribution Model Based on Seesaw Model

The relationship among many subsystems in multi-agent complex systems is difficult to quantify and the coordination among multiple processes in multivariate complex processes is hard to clear analysis. In order to solve the problems this paper presents a seesaw model for the basic dual relation of the complex systems and complex processes. The seesaw model can be applied in various fields and in all aspects of people’s lives. In this paper, the application of this model is derived, and the pre-distribution model of distribution industry is obtained. We analyze the time factor of the distribution and optimize the distribution process by using the seesaw model. Concorde process makes the dissatisfaction degree of distribution service decreased. Pre-distribution makes that delivery speed and delivery efficiency are improved and ensures that dissatisfaction degree of distribution service is effectively reduced.

Mingshan Xie, Yanfang Deng, Yong Bai, Mengxing Huang, Zhuhua Hu

An Efficient Filtration Method Based on Variable-Length Seeds for Sequence Alignment

With the rapid development of next-generation sequencing (NGS) platforms, more than billions of reads are produced quickly. Finding all mapping locations of these reads in the reference genome is not only a bioinformatics issue, but also a large-scale computation issue. Existing all mapping tools are usually divided into the two steps, filtration and verification. Filtration step discards some wrong locations and generates candidates. As for verification step, each candidate is mapped to the reference sequence to determine whether it is a mapping location. Statistics indicated that the verification step is the main part of the whole mapping time. That is to say, less candidates lead to less mapping time. Our strategies improve filtration step to decrease the number of candidates. We propose a dynamic programming and two heuristic strategies and integrated them into the filtration step. These strategies are applied in the state-of-the-art all-mapper, Bitmapper. Compared with the advanced all-mappers, experiment results show that our method make a significant progress.

Ruidong Guo, Haoyu Cheng, Yun Xu

An Optimized Fusion Method for Double-Wearable-Wireless-Band Platform on Cloud-Health Application

This paper presents a stable double-wireless-wearable-band platform that can detect hand gestures. The real-time monitoring and control system utilizes an MCU processor, a wireless transceiver, and a commercial three-axis, digital-output MEMS accelerometer. To detect the user’s hand movements, a 3D virtual environment is created via a double-wearable-band controller. Compared with a single wearable band, double wearable bands can identify more gestures with improved stability. Performances in terms of control and detection are discussed in detail. This research development allows the user to specify desired two-hand postures using the multi-sensor information fusion technique for controlling a variety of robotic devices. In the system, the defined two-hand postures also allow the user to add freestyle control to various applications, which bridge the communication gap between humans and the systems. Moreover, the integration of the action recognition algorithm of the combination of two bracelets and the server brings out a real-time approach to analyze and make decisions based on the users’ data. Therefore, the system can call for help in a timely manner under critical conditions.

Wenchao Xu, Yanbo Liu, Yanqin Yang, Xiaoshuang Ning, Tianxing Chu, Hongzhi Song

Research on Concept Drift Detection for Decision Tree Algorithm in the Stream of Big Data

With the rapid development of information technology, various industries have to deal with an increasing number of data. Compared with the traditional static data, stream data under big data environment was rapid, continuous and always changed with time. At the same time, the implicit distribution of data stream brought about the concept drift. A stream data concept drift detection algorithm named ADDS (Anti-concept Drift Detection Algorithm) was put forward, which is mainly used to detect and process the hidden concept drift of unsteady data stream, under big data environment. The ADDS was focused on the improvements of traditional classification algorithms with incremental way to adapt to the demand of streaming data processing. The experimental results showed that the ADDS had a better concept drift detection effect.

Shangdong Liu, Lili Lu, Yongpan Zhang, Tong Xin, Yimu Ji, Ruchuan Wang

Review of Various Strategies for Gateway Discovery Mechanisms for Integrating Internet-MANET

Mobile Ad hoc network (MANET) is a short range network which enables various mobile nodes to communicate with each other. The range of mobile node for communication is limited in the MANET so two nodes of different network which are placed in long distance are unable to communicate between them. Utility of MANET can be enhanced by providing Internet connectivity to the mobile nodes. MANET is infrastructureless and Internet is infrastructure type network so they require an interface to connect with each other. So in this situation, Gateway, a device which acts as an interface and a router, provides the MANET nodes connectivity to fixed networks. A mobile node in MANET has to find a route to a gateway first to communicate with the Internet host. It requires an efficient gateway discovery mechanism for improve the overall performance of network. For this process various approaches have been proposed in the past. The popular approaches are reactive, proactive and hybrid gateway discovery mechanisms. This paper focuses on various issues for MANET-Internet integration and their proposed technical solutions by various proposals for different Internet gateway discovery mechanisms to improve the performance.

Lin Yang, Zhijie Han, Rui Wang, YongHang Yan

Research on Extraction Algorithm of Palm ROI Based on Maximum Intrinsic Circle

Aiming at the problem that using maximum inscribed circle algorithm to extract the region of interest (ROI) in palm vein image is retrieved and recorded too much, an improved maximum inscribed circle algorithm is proposed. First of all, the palm vein image was done by background separation and smoothing filter preprocessing. Secondly, the grids were added for the preprocessed image, the selection range of the center was determined by the assistant of grids, the initial radius and the variable value of radius were defined. Then, changed the radius in variable size, recorded all inscribed circles. Finally, compared and obtained the maximum inscribed circle of the palm vein image. The circle was the ROI which was needed. The experimental results show that the execution time of the new algorithm is reduced by 10.7 ms, 10.2 ms, 11.3 ms and 10.8 ms under four groups of samples respectively. Therefore, the new algorithm effectively improves the efficiency of the maximum inscribed circle algorithm.

Gang Liu, Jing Zhang

Power Adaptive Routing Scheme with Energy Hole Avoidance for Underwater Acoustic Networks

Underwater Wireless Sensor Networks (UWSNs) have grown rapidly in recent decade. As acoustic communication consumes much more power and underwater deployment environments are much harsher than that of the terrestrial wireless sensor networks, energy efficiency is much more critical for UWSNs. Therefore, it is essential to deal with the energy hole problem and balance the power consumption in underwater acoustic network. In this article, we propose a Power Adaptive and Energy Balance routing scheme (PAEB) to avoid the energy holes taken place owing to the unbalanced energy consumption in UWSNs. In order to balance the energy consumption, a Binary Exponential Transmission (BET) scheme is introduced. In BET scheme, sensor node located in different layer may have different transmission distance according to its residual energy. When sensor node sends packet to the sink, it selects the optimal transmission radius and adjusts the transmission power based on its current power level. The much energy is residual, the further the transmission radius is selected. The experimental results show that PAEB performs better than the existing representative protocols in terms of network lifetime and end-to-end delay.

Xian-yi Chen, Guo-lan Lin

A Lightweight Algorithm for Computing BWT from Suffix Array in Disk

The Burrows-Wheeler transform (BWT) and the suffix array (SA) of an input string are important data structures widely used in modern bioinformatics researches such as full-text search, alignment etc. In this paper, we present a lightweight external memory algorithm for computing the BWT from a given suffix array and the input string. The algorithm has a linear I/O complexity O(n) and a workspace of at most n/2 integers. An experiment study is conducted to evaluate the time and space performance of the proposed algorithm on a number of realistic datasets. The experimental results are consistent with the theoretical complexities of the algorithm.

Jing Yi Xie, Bin Lao, Ge Nong

H ∞ Filtering in Mobile Sensor Networks with Missing Measurements and Quantization Effects

This paper is concerned with the $$ {{H}_{\infty}} $$ filtering problem for an array of 2D distributed parameter systems over lossy mobile sensor networks. The mobile sensor network suffers from missing measurements as well as quantization effects that are presented in a new framework. Bernoulli distribution is introduced to govern the data missing. A new $$ {{H}_{\infty}} $$ filtering technique is proposed for the addressed 2D semi-linear parabolic systems. Sufficient conditions are established in terms of some inequalities and the velocity law of each mobile sensor, such that the filtering error system is globally asymptotically stable in the mean square and has a guaranteed prescribed disturbance attenuation level $$ \gamma $$ for all nonzero noises. Finally, a numerical example is exploited to show the effectiveness of the proposed filtering scheme.

Xueming Qian, Baotong Cui

A Cost-Effective Wide-Sense Nonblocking k-Fold Multicast Network

Multicast is one of the most dense communication patterns. Any destination node of a k-fold multicast network can be involved in up to k simultaneous multicast connection. The hardware cost of traditional k-fold switching network for wide-sense nonblocking multicast is typically very high. In this paper, we propose a new wide-sense nonblocking k-fold multicast network and multicast routing algorithm. The k-fold design has significantly lower network cost than that of k copies of 1-fold multicast networks. The time complexity of the corresponding routing algorithm is no higher than that of previous works.

Gang Liu, Qiuming Luo, Cunhuang Ye, Rui Mao

The Design of General Course-Choosing System in Colleges and Universities

The college credit system allows students to take their own courses according to their own interests and development plans, but since the time for course-choosing is limited and the huge corresponding data requires a large amount of human resources, it is necessary to use computer software for course-choosing. The proposed system follows the complete design flow of computer software. System requirement analysis was done around the core function of the system which is course selection. Based on the system requirement analysis, the system function and the data structure were determined. And then, the system information flow, the code generation algorithm, the back-end database, the course setting algorithm, and the course-choosing algorithm were designed. In addition, a new algorithm for student ID generation is proposed. The student ID generated by this algorithm has simple structure, and also follows the principle of unique, reasonableness and extensibility. Therefore, this course-choosing system could be applied in every college and university.

Chunmin Qiu, Shaojie Du, Bailu Zhao

Research on Vectorization Technology for Irregular Data Access

Current program vectorization methods support continuous memory access forms. There are few researches on vectorization of irregular data access. In order to improve the program execution efficiency, vectorization technology of irregular data access is researched. A vectorization method for non-continuous access and indirect array indexes is proposed. This paper designs a calculation method of vectorization performance gains and analyzes the different performance gains of different non-continuous access vectorization methods. Finally the experimental results show that this method can vectorize irregular data access effectively and improve the program execution efficiency.

Wang Qi, Han Lin, Yao Jinyang, Liu Hui

A New Simple Algorithm for Scrambling

The paper proposes a new algorithm which use simple divisible and xor operation to change the values of pixels and the site of pixels in digital images based on logistic system. This algorithm encrypted images successfully. Experiments show that the algorithm is simple and easy to do. It has a large secret-key space and higher-security. Furthermore, the algorithm not only has preferable practicability, but also can resists various stacks.

Xing Zeng, Xiulai Li, Yali Luo, Mingrui Chen

The Framework of Relative Density-Based Clustering

Density-based clustering, using two-phase scheme which consists of an online component and an offline component, is an effective framework for data stream clustering, it can find arbitrarily shaped clusters and capture the evolving characteristic of real-time data streams accurately. However, the clustering has some deficiencies on offline component. Most algorithm don’t adapt to the unevenly distributed data streams or the multi density distribution of the data streams. Moreover, they only consider the density and centroid to connect the adjacent grid and ignore similarity of attribute value between adjacent grids. In this paper, we calculate the similarity of neighboring grids and take the similarity as a weight that affects the connection of the neighboring grids and propose the relative density-based clustering that cluster the grids based on relative difference model that considers the density, centroid and the weight of similarity between adjacent grids, simply, we connect neighboring grids which are the relative small difference to form clusters on offline component. The experimental results have shown that our algorithm apply to the unevenly distributed data streams and has better clustering quality.

Zelin Cui, Hong Shen

Efficient Algorithms for VM Placement in Cloud Data Center

Virtual machine (VM) placement problem is a major issue in cloud data center. With the rapid development of cloud computing, efficient algorithms are needed to reduce the power consumption and save energy in data centers. Many models and algorithms are designed with an objective to minimize the number of physical machines (PMs) used in cloud data center. In this paper, we take into account the execution time of the PM, and formulate a new optimization problem of VM placement, which aims to minimize the total execution time of the PMs. We discuss the NP-hardness of the problem, and present heuristic algorithms to solve it under both offline and online scenario. Furthermore, we conduct experiments to evaluate the performance of the proposed algorithms and the result show that our methods are able to perform better than other commonly used algorithms.

Jiahuai Wu, Hong Shen

Weighted One-Dependence Forests Classifier

Averaged One-Dependence Estimators (AODE) combines all Super Parent-One-Dependence Estimators (SPODEs) with ensemble learning strategy. AODE demonstrates good classification accuracy with very little extra computational cost. However, it ignores the dependences between attributes. In this paper, we propose aggregating extended one-dependence estimators named Weighted One-Dependence Forests (WODF) which splits each SPODE into multiple subtrees by attribute selection. WODF assigns the weight to every subtree with conditional mutual information. Extensive experiments and comparisons on 40 UCI data sets demonstrate that WODF outperforms AODE and state-of-the-art weighted AODE algorithms. Results also confirm that WODF provides an appropriate tradeoff between runtime efficiency and classification accuracy.

Guojing Zhong, Limin Wang

Research and Realization of Commodity Image Retrieval System Based on Deep Learning

This paper proposed a commodity image retrieval system based on CNN and ListNet sort learning method. CNN contained two convolutional layers, two pooling layers and two innerproduct layers. ReLu function was used as the activation function after the convolutional layer, achieving the sparsity and preventing the disappearance of the gradient. The pooling layer used stochastic pooling method and improved the generalization ability of the model. In addition, softmax regression was used for classification. Innerproduct layer adopted dropconnect method, which is more powerful than the generalization of dropout, and it can effectively prevent the occurrence of the overfitting. What’s more, the feature extraction of the network was optimized by stochastic gradient descent (SGD) algorithm. And we combined the learn to rank algorithm of the text retrieval domain. We used ListNet algorithm to combine a variety of feature vectors, solving the problem of the image retrieval.

Cen Chen, Rui Yang, Chongwen Wang

An Improved Algorithm Based on LSB

LSB algorithm is a common image information hiding method. Based on Logistic chaotic sequence, an improved LSB algorithm is proposed. With the XOR operation to complete the encryption of embedded information. The simulation results show that the algorithm has good safety performance and can resist the pollution of salt and pepper noise. It is a kind of practical algorithm.

Yali Luo, Xiulai Li, Chaofan Chen, Mingrui Chen

A Report on the Improvement of Information Technology Capability of Teachers in Primary and Middle Schools in Hainan Province

With the rapid development of our technology, information technology capacity has become the necessary professional ability for teachers. In June 2014, the Ministry of Education promulgated the “primary and secondary school teacher information technology application ability standard (Trial)”, opened a new round of primary and secondary school teachers in information technology application ability improve training. In order to understand the current level of primary and secondary school teachers in application of information technology capabilities in Hainan for the implementation of capacity training to provide the basis. This paper take 856 primary and middle school teachers in Hainan Province as the research object, and carries out a more comprehensive research on the status quo of its application technology from the standard angle.

JingYu Luo, Zhao Qiu, JianZheng Hu, XiaWen Zhang

The Research of the Airport Retail Layout Based on the Location Model

With the airport authorities realizing the importance of the non-aviation revenue, the airport retails are taken as one of the most important non-aviation business, how to give it a reasonable layout is becoming the most important point for the airport commercial plan. Firstly, it set the model hypothesis considering the characteristics and structure of airport terminal, and carried out the location model of airport model. Secondly, the airport retails shops of the landside hall and airside hall were relayouted and improved the efficiency and productivity of retails shops. Finally, it drew a conclusion that airport layout of airport retail is very complex, layouting the airport retail using airport retail location model is efficient but should think about safety policies, passenger procedure and passenger number further.

Han-tao Yang

Research on Model and Method of Relevance Feedback Mechanism in Image Retrieval

Considering the limitations of image retrieval method based on computer center, the introduction of relevance feedback mechanism increasingly shows its importance in image retrieval. This paper makes a deep research on some commonly used relevance feedback models and feedback methods in image retrieval. The purpose is to improve the query efficiency and retrieval precision of image retrieval.

Xinying Li, Taijun Li, Feng Li, Hongli Wu

Processing Redundancy in UML Diagrams Based on Knowledge Graph

UML (Unified Modeling Language) is designed for all stages of software development, but it lacks precise semantic information. MDE (Model-driven engineering) takes the model as the primary software product and its main research direction is modeling and model transformation. We propose to explore the entity abstraction scenario where no existing classes fit as the representative entity for other classes through the data, information and knowledge recreation of existing classes and relationships with the introduction of knowledge graphs. In this paper, we proposed a knowledge graph to enhance model design between models.

Yirui Jiang, Yucong Duan, Mengxing Huang, Mingrui Chen, Jingbin Li, Hui Zhou

An Automatic Fall Detection System Based on Derivative Dynamic Time Warping

Maturation of Internet and rapid development in mobile communication make smart device could bring enhanced services to person especially in health care center. Therefore, we focus on developing a fall detection application running on Android mobile phones. The detector is fit to indoor and outdoor, without confine of its surroundings. We propose a novel method which fuses Derivative Dynamic Time Warping (DDTW) to detect a fall event and algorithm sensitivity is 84.7%, as well as 94% of specificity. Our algorithm is considerable concise and efficient, what’s more, it do not intrude on privacy of its users or degrade the quality of life. And above all, the method not only overcomes the shortage of thresholding-based fall detection method, but also applicable to all kinds of people with different weight and height.

Hong Yang, Yanqin Yang, Wenchao Xu, Yuxin Pang

On Signal Timing Optimization in Isolated Intersection Based on the Improved Ant Colony Algorithm

The unreasonable allocation of traffic lights is liable to trigger traffic jam. The ant colony algorithm is a universal random optimization algorithm, which can solve the problem of traffic signal timing. Aiming at the problem of signal timing optimization in isolated intersection, the paper comes up with the improved ant colony algorithm, which is showed by the experimental simulation results that it is better than the traditional way in signal timing.

Huang Min

Experiments on Neighborhood Combination Strategies for Bi-objective Unconstrained Binary Quadratic Programming Problem

Local search is known to be a highly effective metaheuristic framework for solving a number of classical combinatorial optimization problems, which strongly depends on the characteristics of neighborhood structure. In this paper, we integrate the neighborhood combination strategies into the hypervolume-based multi-objective local search algorithm, in order to deal with the bi-objective unconstrained binary quadratic programming problem. The experimental results show that certain combinations are superior to others. The performance analysis sheds lights on the ways to further improvements.

Li-Yuan Xue, Rong-Qiang Zeng, Wei An, Qing-Xian Wang, Ming-Sheng Shang

Porting Referential Genome Compression Tool on Loongson Platform

With the fast development of genome sequencing technology, genome sequencing become faster and affordable. Consequently, genomic scientists are now facing an explosive increase of genomic data. Managing, storing and analyzing this quickly growing amount of data is challenging. It is desirable to apply some compression techniques to reduce storage and transferring cost. Referential genome compression is one of these techniques, which exploited the highly similarity of the same or an evolutionary close species (e.g., two randomly selected humans have at least 99% of genetic similarity) and store only the differences between the compressed file and well-known reference genome sequence. In this paper, we port two referential compression algorithm to Loongson platform and profiling their performance. And we use multi-process technology to improve the speed of compression.

Zheng Du, Chao Guo, Yijun Zhang, Qiuming Luo

Statistics of the Number of People Based on the Surveillance Video

With the rapid development of economy, large-scale entertainment, sporting events, religious ceremonies and other major activities, there are often a large crowd gathered. In order to avoid accidents occurred, counting the number of development becomes more and more important. And in the daily activities of life relatively slim, for example: the school attendance rate statistics, the convenience given by the statistics of pedestrians should not be underestimated. The number of statistical methods is drawing more and more people’s attention.

Zhao Qiu, ShiYao Lei, JianZheng Hu, JingYu Luo

Differential Privacy in Power Big Data Publishing

In recent years, the development of smart grid leads to the explosive growth of power big data. Power companies can analyze these data to provide personalized services to users. However, the analysis of power big data can have the risk of user privacy disclosure. The performance of the traditional algorithms is not satisfied due to the complexity of the power big data on preventing information leakage. Distributed and heterogeneous data generated in the operation, maintenance and other processes of electricity smart grid can cause the complexity of the data. This paper proposes a method of differential privacy to preserve privacy in the power big data publishing. The experimental results shows that the performance of our method is convincing.

Ping Kong, Xiaochun Wang, Boyi Zhang, Yidong Li

Parallel Architectures


Parallel Aligning Multiple Metabolic Pathways on Hybrid CPU and GPU Architectures

Metabolic pathway alignment remains an important tool in systems biology, and has become even more important with the growing mass of metabolic pathway data. However, the process of aligning multiple metabolic pathways scale poorly with either the number of pathways or with the size of pathways, and no attempts have been made to exploit the parallelism of the pathway alignments to improve the efficiency. This paper proposes a parallel metabolic pathway alignment method called PMMPA. In PMMPA, we design a commonly used parallel algorithm for the computation of reaction (node) similarity in GPU, and implement a parallel strategy for aligning multiple metabolic pathways in multi-core CPU. The experimental results show that this parallel alignment implementation achieves at most 300 times faster than the single-threaded version, the parallel implementation of aligning metabolic pathways on the hybrid CPU and GPU architecture is promising in improving the efficiency.

Yiran Huang, Cheng Zhong, Jinxiong Zhang, Ye Li, Jun Liu

Speeding Up Convolution on Multi-cluster DSP in Deep Learning Scenarios

Recently, deep learning has achieved great success in artificial intelligent, whose superiority also brought new opportunity for the related research in embedded system. This paper focused on optimizing and speeding the convolution computing, the core operation within convolution neural network based on a multi-cluster digital signal processor, BWDSP. By taking advantage of the BWDSP’s architecture and characteristics of convolution computation, a suitable parallel algorithm was designed. Based on features of convolution neural network model structure, an automatic optimization tool for convolution computing with specific arguments was presented as well. The experimental result showed that the parallel algorithm given in this paper is 9.5x faster than GEMM-based algorithm commonly used in GPU and 5.7x faster than the traditional vectorization optimization algorithm. Meanwhile, a comparison was made between the parallel algorithm and tiled-base algorithm widely adopted in system with cache hierarchies, showing that the parallel one could achieve a better performance density of 1.55 times than that of later one, meaning that the work in this paper can make full use of computing resources to make them more efficient.

Deng Wenqi, Yang Zhenhao, Lu Maohui, Wang Gai, Yang JiangPing, Zheng Qilong

Optimization Scheme Based on Parallel Computing Technology

Parallel computing is a high performance technology to solve problems, in order to improve computing efficiency, we use the processor to concurrent execute several parts divided from one problem. Based on the current issues in parallel computing area, both the data processing repetition rate and the parallel computing time depend on the time of the last thread in the task completing. This paper was written to take an overview of the existing parallel computing techniques and structures, and propose a solution of adding an advanced thread or advanced processor to make up the deficiency in parallel computing area.

Xiulai Li, Chaofan Chen, Yali Luo, Mingrui Chen

Research on Client Adaptive Technology Based on Cloud Technology

With the highly popularity of the Internet, the diversification of the Internet terminal equipment and applications, client development technology and the operation and maintenance of the architecture, development costs and performance are facing more and more challenges. This paper introduces the Web technology in the cloud environment, analyzes the present client adaptive technology and existing problems, put forward a kind of adaptive client cloud environment model, based on Response Web Design, combined with multi template and backend For Frontends (BFF), based on cloud technology, put forward the template fragment combination and on-demand production principle, To solve the adaptive problem of client diversification and intelligent; saving development costs and improve development efficiency.

Xiaojing Zhu

Resource Allocation and Energy Management Based on Particle Swarm Optimization

In recent years, resource allocation and management has become a hot topic among people. Home resource’s allocation and management is still a problem when home electricity power as a resource, because the resource control of the load is difficult to achieve. To solve these problems, we must implement a resource allocation and management system for home resource. Particle Swarm Optimization (PSO) algorithm is an effective approach to solve the optimal problem. In this paper, a resource allocation and energy management system (RAAEMS) is proposed to support the resource control based on real-time for home energy resource, and the objective function is solved by using PSO algorithm.

Gang Mei, Mingrui Chen, Xing Zhen

A Parallel Clustering Algorithm for Power Big Data Analysis

With the fast development of information technology, the power data is growing at an exponentially speed. In the face of multi-dimensional and complicated power network data, the performance of the traditional clustering algorithms are not satisfied. How to effectively cope with the power network data is becoming a hot topic. This paper proposes a parallel implement of K-means clustering algorithm based on Hadoop distributed file system and Mapreduce distributed computing framework to deal this problem. The experimental results show that the performance of our proposed algorithm significantly outperforms the traditional clustering algorithm and the parallel clustering algorithm can significantly reduce the time complexity and can be applied in analyzing and mining of the power network data.

Xiangjun Meng, Liang Chen, Yidong Li

Customized Filesystem with Dynamic Stripe Strategies on Lustre-Based Hadoop

With large-scale data exploding so quickly that the traditional big data processing framework Hadoop has met its bottleneck on data storing layer. Running Hadoop on modern HPC clusters has attracted much attention due to its unique data processing and analyzing capabilities. Lustre file system is a promising parallel storage file system occupied HPC file system market for many years. Thus, Lustre-based Hadoop platform will pose many new opportunities and challenges on today’s data era. In this paper, we customized our LustreFileSystem class which inherits from FileSystem class (inner Hadoop source code) to build our Lustre-based Hadoop. And to make full use of the high-performance in Lustre file system, we propose a novel dynamic stripe strategy to optimize stripe size during writing data to Lustre file system. Our results indicate that, we can improve the performance obviously in throughput (mb/sec) about 3x in writing and 11x in reading, and average IO rate (mb/sec) at least 3 times at the same time when compared with initial Hadoop. Besides, our dynamic stripe strategy can smooth the reading operation and give a slight improvement on writing procedure when compared with existing Lustre-based Hadoop.

Hongbo Li, Yuxuan Xing, Nong Xiao, Zhiguang Chen, Yutong Lu

Research on Optimized Pre-copy Algorithm of Live Container Migration in Cloud Environment

Some load imbalance problems emerge under the distributed cloud computing platform based on container. And it is necessary to transfer the container in the higher load server to another relatively idle servers. The traditional pre-copy algorithm ignores the characteristics of the memory pages, which causes memory pages with high re-modified rate copying many times in the iterative copy phase. Therefore, this paper proposes an optimized pre-copy algorithm (OPCA) which introduces a Gray-Markov prediction model. Memory pages are added with high re-modified rate into the Hot Workspace by the prediction model, and these memory pages are involved in the stop-copy. The experiments show that OPCA reduces the number of iterations and downtime. Moreover, it also improves the utilization of resources and enhances the user’s experience.

Huqing Nie, Peng Li, He Xu, Lu Dong, Jinquan Song, Ruchuan Wang

A Cost Model for Heterogeneous Many-Core Processor

Heterogeneous many-core processors become an important trend in high-performance computing area, but their sophisticated architecture greatly complicates the programming and compiling issue. The cost model is an important part of optimizing compilers, which is used to analyze the benefits of various program optimizations. This paper constructs a cost model for SW26010 heterogeneous many-core processor, and proposes a dynamic-static hybrid method to analyze benefit based on this cost model. Then these have been implemented in an automatic parallelizing framework for SW26010. The experimental results show that the cost model and the benefit analysis can filter a large number of non-beneficial parallel loops and the performance of the automatically parallelized programs increases significantly.

Yanbing Li, Qi Wang, Yingying Li, Lin Han, Yuchen Gao, Qing Mu

Scalable K-Order LCP Array Construction for Massive Data

Given a size-n input text T and its suffix array, a new method is proposed to compute the K-order longest common prefix (LCP) array for T, in terms of that the maximum LCP of two suffixes is truncated to be at most K. This method employs a fingerprint function to convert a comparison of two variable-length strings into a comparison of their fingerprints encoded as fixed-size integers. This method takes $$ {\text{O}}\left( {n\,\log K} \right) $$ time and $$ {\text{O}}\left( n \right) $$ space on internal and external memory models. It is also scalable for a typical distributed model consisting of $$ d $$ computing nodes, where the time and space complexities are evenly divided onto each node as $$ {\text{O}}\left( {n\,\log K/d} \right) $$ and $$ {\text{O}}\left( {n/d} \right) $$, respectively. For performance evaluation, an experimental study has been conducted on both external memory and distributed models. From our perspective, a cluster of computers in a local area network is commonly available in practice, but there is currently a lack of scalable LCP-array construction algorithm for such a distributed model. Our method provides a candidate solution to meet this demand.

Yi Wu, Ling Bo Han, Wai Hong Chan, Ge Nong

A Load Balancing Strategy for Monte Carlo Method in PageRank Problem

PageRank algorithm is key component of a wide range of applications. Former study has demonstrated that PageRank problem can be effectively solved through Monte Carlo method. In this paper, we focus on efficiently parallel implementing Monte Carlo method for PageRank algorithm based on GPU. Aiming at GPU, a parallel implementation must consider instruction divergence on the single instruction multiple data (SIMD) compute units. Due to the fact that low-discrepancy sequences are determined sequences, we adopt the low-discrepancy sequences to simulate the random walks in PageRank computations in our load balancing strategy. Furthermore, we allocate each thread of a block to compute a random walk of each vertex with a same low-discrepancy sequence. As a result, no idle thread exists in the PageRank computations and warp execution efficiency is up to 99%. Moreover, our strategy loads the low-discrepancy sequences into shared memory to reduce the data fetch cost. The results of experiments show that our strategy can provide high efficiency for Monte Carlo method in PageRank problem in GPGPU environment.

Bo Shao, Siyan Lai, Bo Yang, Ying Xu, Xiaola Lin

Experiences of Performance Optimization for Large Eddy Simulation on Intel MIC Platforms

Large Eddy Simulation (LES) is a mathematical model for turbulence used in Computational Fluid Dynamics (CFD). We have implemented LES on multi-core CPUs and General Purpose Graphics Processing Units (GPGPUs). In this work, we port and optimize LES on Intel Many Integrated Core (MIC) platforms. On Intel MIC co-processor (KNC), we implement LES using the main execution modes, including native, offload and symmetric execution modes. The newly emerging second generation of Intel MIC processor (Knights Landing, i.e. KNL) acts as an independent multi-core computing node, it is more convenient to port the application. On both of the MIC platforms, some important performance optimization techniques are implemented and evaluated, such as parallelization with OpenMP threads and MPI processes, single-instruction-multiple-data (SIMD) vectorization, memory access optimization, threads scheduling, etc. The experimental results demonstrate that performance optimization techniques are very important when porting applications on MIC platforms.

Zhengxiong Hou, Chengwen Zhong, Christian Perez, Qing Zhang, Yunlan Wang


Weitere Informationen

BranchenIndex Online

Die B2B-Firmensuche für Industrie und Wirtschaft: Kostenfrei in Firmenprofilen nach Lieferanten, Herstellern, Dienstleistern und Händlern recherchieren.



Globales Erdungssystem in urbanen Kabelnetzen

Bedingt durch die Altersstruktur vieler Kabelverteilnetze mit der damit verbundenen verminderten Isolationsfestigkeit oder durch fortschreitenden Kabelausbau ist es immer häufiger erforderlich, anstelle der Resonanz-Sternpunktserdung alternative Konzepte für die Sternpunktsbehandlung umzusetzen. Die damit verbundenen Fehlerortungskonzepte bzw. die Erhöhung der Restströme im Erdschlussfall führen jedoch aufgrund der hohen Fehlerströme zu neuen Anforderungen an die Erdungs- und Fehlerstromrückleitungs-Systeme. Lesen Sie hier über die Auswirkung von leitfähigen Strukturen auf die Stromaufteilung sowie die Potentialverhältnisse in urbanen Kabelnetzen bei stromstarken Erdschlüssen. Jetzt gratis downloaden!