Loading...
  • 中国计算机学会会刊
  • 中国科技核心期刊
  • 中文核心期刊

Current Issue

    • Porting and optimization of GROMACS 2020 on ROCm platform
      ZHANG Yu-zhou, CAO Wu-di, BU Jing-de, TAN Guang-ming, JI Qing
      2021, 43(11): 1901-1909. doi:
      Abstract ( 236 )   PDF (1090KB) ( 250 )     
      GROMACS is a widely used open-source molecular dynamics simulation software. Currently, NVIDIA GPUs are mainly used for accelerated calculations through CUDA. ROCm is an open-source high-performance heterogeneous computing platform. Based on the HIP programming language of the ROCm platform, this paper implements the complete porting of the GROMACS 2020 series on the ROCm platform for the first time. On MI50 GPU, with a complex ionic liquid simulation example as the target, the performance analysis of the transplanted code was carried out using GPU performance analysis tool rocprof. According to the hardware characteristics of MI50, the bonding force kernel function, the PME kernel function of electrostatic force and the short-range non-bonding force kernel function are optimized successively. After optimization, the performance of the target calculation example is about 2.8 times that of the initial version. The performance on MI50 is 60.5% higher than that of the GROMACS original OpenCL code, which is about 2.7 times faster than the pure CPU version. In the single-node test of the other two representative examples and the multi-node scalability test of the ionic liquid example, the optimized code also achieves a better performance improvement, which shows that the optimization has a certain versatility. 

      A new multi-classification task accuracy evaluation method based on confusion matrix
      ZHANG Kai-fang, SU Hua-you, DOU Yong
      2021, 43(11): 1910-1919. doi:
      Abstract ( 265 )   PDF (531KB) ( 281 )     
      The accuracy evaluation of multi-classification tasks has important theoretical significance and application value to the classification effect of the evaluation model. Aiming at multi-classification tasks in the field of machine learning, based on the current existing methods, this paper proposes a new method by expanding and migrating applications. In order to accurately evaluate the classification effect of the multi-classification task model, this paper introduces the remote sensing image classification effectiveness evaluation method (R′)  into the multi-classification tasks. In view of the actual characteristics of the multi-classification tasks, the method improves and popularizes the R′ method to better evaluate the performance of classifiers. The experimental results on the recognition task of MNIST handwritten character set and the classification task of the CIFAR-10 dataset show that, although the calculation is also based on the confusion matrix, compared with the existing evaluation indicators, the method can simultaneously give the overall classification performance of the classifiers and the classification efficiency of the individual categories, which can be beneficial to the training process. On the other hand, the method can be extended to the classification performance evaluation of any classification tasks, which has a good application prospect.

      A parallel predication method of underwater acoustic environment characteristic parameters
      FAN Pei-qin, GUO Wu-hong, HAN Mei, TANG Shuai, Zhang Chi,
      2021, 43(11): 1920-1925. doi:
      Abstract ( 109 )   PDF (710KB) ( 136 )     
      With the rapid development of underwater acoustic equipment, the coupling between its performance and ocean environment is becoming stronger and stronger. How to provide long-time, wide-ranger and fine underwater acoustic characteristic parameter information for underwater acoustic sensors is of great importance for optimizing the design of underwater acoustic sensors and making full use of their detection performance to realize the best matching between ocean environment and sensor performance. The parallel prediction program of underwater acoustic environment characteristic parameters is developed by using the MPI parallel programming environment. Aiming at the problem of uneven load distribution in parallel programs, the causes of uneven load distribution are analyzed, the strategies and methods of performance optimization are given. The results show that the load balancing problem of the optimized parallel program is improved, the computing time is greatly reduced, and the prediction ability of underwater acoustic environment parameters is greatly improved.

      Package dependency analysis for operating system distribution building
      MA Jun, ZHOU Kai, REN Yi, ZHU Hao, QIN Ying, WANG Jing
      2021, 43(11): 1926-1933. doi:
      Abstract ( 184 )   PDF (712KB) ( 293 )     
      Linux-based open-source operating system distributions are usually built from a series of interrelated packages. Due to the large number of packages, complex relationships such as dependencies and conflicts, the operating system distribution building is also complicated. The granularity and precision of customization are insufficient, and there may be many redundant packages. The current operating system distribution building is mainly organized through engineering experience, and especially the selection of software packages lack of theoretical analysis and guidance. This paper proposes a basic model of operating system distribution construction based on dependency relationship. Combined with the distribution building process and software repository data of Ubuntu, statistical analysis and model verification are carried out on the characteristics of software packages, such as in_degree and out_degree, priority, classification, etc. This paper summarizes the main principles of the operating system distribution building from the perspective of dependency relationship, and provides guidance and important reference for the subsequent improvement of the precision of automatic distribution customization, software release evolution and bug analysis.

      A test time optimization algorithm for multi-tower 3D SoCs based on partially pipelined test
      A test time optimization algorithm for multi-tower D SoCs based on partially pipelined test
      2021, 43(11): 1934-1943. doi:
      Abstract ( 85 )   PDF (1435KB) ( 109 )     
      Aiming at the mid-bond test of hard-die based multi-tower 3D SoCs, this paper proposes a novel test time optimization algorithm considering the test resource constraints such as test access mechanism, Through-Silicon-Via count, and test power consumption. Once any remaining test resources do not satisfy the requirements of the die to be scheduled for testing, the test resources for the die that ends testing the earliest are released, until the die to be scheduled can be tested ahead of time as much as possible, thus obtaining partially pipelined testing between the newly scheduled die and the unfinished die. Five typical circuits in the ITC02 test benchmark circuit were selected, and two types of multi-tower 3D SoCs containing sub-towers were manually constructed. The results show that, compared with the existing algorithms, the proposed algorithm reduces the idle time blocks and significantly shortens the total test time. In addition, compared with increasing the number of TSVs, increasing the number of test pins can effectively reduce the total test time of a multi-tower 3D SoC. 

      WSN area coverage optimization based on Delaunay triangulation strategy
      ZHANG Jing, WEI Miao,
      2021, 43(11): 1944-1951. doi:
      Abstract ( 134 )   PDF (1142KB) ( 156 )     
      Aiming at the problem that traditional geometric methods are difficult to apply to probabilistic perception models in the process of secondary deployment of nodes in wireless sensor networks, a WSN area coverage optimization scheme based on Delaunay triangulation strategy is proposed. Firstly, Delaunay triangulation is performed on randomly scattered static nodes and the edge vertices in the monitoring area to obtain a static node triangulation. Combined with the probabilistic perception model, it is proved that there are completely uncovered areas inside triangles. Secondly, the selected triangle centroid set is used as the initial solution set of the particle swarm optimization algorithm, and the improved particle swarm optimization algorithm is used to complete the secondary deployment of mobile nodes to achieve the purpose of repairing the coverage loophole. The simulation experiment proves that this optimization scheme can effectively repair the coverage loopholes and significantly improve the regional coverage.

      Difficulties and solutions of computer network management
      WANG Hong, WANG Cheng-song, LI Su-dan
      2021, 43(11): 1952-1958. doi:
      Abstract ( 111 )   PDF (585KB) ( 152 )     
      With the advent of the 5G era, the rapid deployment capabilities of network services and the requirements for network operation and maintenance capabilities pose new challenges to network management. On the one hand, the network is undergoing a network function virtualization transformation, and slicing and microservices complicate the network. On the other hand, network managers need a simpler and automated tool set to support on-demand, real-time, and flexible network services. This article analyzes the current difficulties faced by network management and the reasons for the difficulties, and proposes a research framework for network self-management to provide references for further research. 

      An improved remote user authentication scheme
      CAO Shou-qi, HE Xin, LIU Wan-rong
      2021, 43(11): 1960-1965. doi:
      Abstract ( 114 )   PDF (653KB) ( 157 )     
      The rapid development of the Internet of Things has brought great convenience to peoples production and life, but how to ensure the information security of users is an important issue that must be solved in the development of the Internet of Things. In order to solve this problem, researchers must propose a more secure identity authentication protocol under the premise of increasing the amount of computation. Nikooghadam et al. proposed a protocol to protect user identity. This paper analyzes it and finds some disadvantages. Based on the protocol of Nikooghadam et al., an improved remote user authentication protocol is proposed. BAN logic is adopted to verify the protocol. Performance comparison and computational efficiency analysis are also carried out, and the results show that the proposed protocol has higher security under the premise of increasing the amount of computation.

      Research and application situation analysis of domestic blockchain technology based on bibliometrics
      PEI Song-ying, CHEN Zhen-guo
      2021, 43(11): 1966-1978. doi:
      Abstract ( 136 )   PDF (1289KB) ( 132 )     
      Abstract: In order to analyze the development trend of blockchain technology, this paper designs the retrieval requirements from three aspects: the overall literature output of blockchain, the application of blockchain and blockchain finance. Taking the literature retrieved from CNKI as a sample, the method of literature econometric analysis is introduced, and CiteSpace software is used to draw the co-occurrence map of key words, the map of author cooperation and institutional cooperation, so as to visually analyze the data. In the analysis, the key words applied in the blockchain are classified according to the meaning, and the relationship between the change speed of the hot key words and the research trend is creatively proposed through the tracking statistics of the blockchain hot key words. The results show that the relevant papers of blockchain have been growing explosively since 2015. The application and research of blockchain in finance, the characteristics of blockchain technology, the combination of blockchain and emerging technology are the hot issues of current research. There is less cooperation between authors and institutions in blockchain research, and scientific research cooperation groups need to be formed. In the future, the research focus of blockchain tends to be the combination of practical application and emerging technology.


      A data-driven software architecture of the measured data’s real-time processing on tracking ships
      ZHANG Yu-xin, LI Yong-gang, SHI Ming-qian, GUO Li-bing, YANG Hai-min, HU Shang-cheng
      2021, 43(11): 1979-1985. doi:
      Abstract ( 110 )   PDF (783KB) ( 153 )     
      The diversity and complexity of space tracking and control tasks continue to increase. With the change and adjustment of data sampling frequency, the original time-driven processing mode has gradually been unable to meet the demand. The data measurement software developed based on the time-driven software architecture is becoming more and more complex, and the difficulty of maintenance is increasing. In response to the above problems, this paper proposes a data-driven software architecture of the measured data’s real-time processing on tracking ships. Under the Kylin operating system, the real-time processing software for external test data is realized based on this architecture. The software is composed of multiple data-driven functional components, and each component uses a service bus to achieve data interaction and integration. Experiments show that the data-driven software of the measured data's real-time process based on this architecture solves the problems of excessive delay and sequence alternations and can meet the requirements of current space tasks.



      Abdominal artery segmentation based on improved convolutional neural network
      JI Ling-yu, GAO Yong-bin, CAI Qing-ping, WEI Zi-ran, LIAO Wei
      2021, 43(11): 1986-199. doi:
      Abstract ( 123 )   PDF (795KB) ( 158 )     
      Abdominal artery segmentation is an essential task for the diagnosis of gastric cancer lymph node metastasis and the judgment of hepatic artery variant type. In order to solve the problems of low segmentation accuracy and easy fracture of abdominal artery, this paper proposes an abdominal artery segmentation method based on improved convolutional neural network. A pre-training module (resnet34) with convolutional attention is employed in encoding part of convolutional network to avoid the disappearance of gradients and better obtain the feature information of the images. In order to expand the receptive field and gather multi-scale feature information, a new multi-scale feature fusion module is proposed. In addition, the learning of the edge structure information of arteries is very significant. Attention guide filtering is introduced as the information expansion path to make the output features more structured and improve the accuracy of vascular segmentation. The proposed method is used to evaluate the performance of the abdominal artery segmentation. The experimental results show that, compared with the basic network U-Net, the sensitivity and intersection-over-union (IOU) of the proposed method are increased by 2.84% and 1.19%, respectively. Compared with the network CE-Net, the sensitivity and IOU are improved by 1.34% and 1.61%, respectively.

      Cubic triangular Bézier surface with shape parameters
      ZHA Dong-dong, LIU Hua-yong, WANG Zeng-zhen
      2021, 43(11): 1994-2002. doi:
      Abstract ( 68 )   PDF (3345KB) ( 91 )     
      In order to improve the shape control capability of the cubic triangular Bézier surface, the Bernstein basis function of the cubic triangular with two parameters is constructed from the point of view of the local and global shape parameters. The cubic triangular  λα-Bézier surface is defined by the basis function, and different control effects are achieved by changing the values of the two parameters. The condition of  C1,G1,  continuity between cubic triangular λα-Bézier surface patches and its proof is given. Related examples also confirm that: the cubic triangular λα-Bézier surface not only inherits the good properties of the cubic triangular Bézier surface, but also improves the shape control ability of the surface by changing the values of the parameters. In surface stitching, the parameter is also changed to construct multiple joining styles.  

      A fundamental matrix estimation method based on improved quasi-affine transformation
      FAN Yi-kai, LIU Shi-jian, PAN Jeng-shyang,
      2021, 43(11): 2003-2010. doi:
      Abstract ( 105 )   PDF (1182KB) ( 109 )     
      On the basis of fundamental matrix estimation, computer vision methods are used to reveal the three-dimensional information of an object within a series of scene images captured from different angles and distances. They are the primary solutions for cutting-edge problems such as image-based modeling and simultaneous localization and mapping. Accuracy and efficiency are two major metrics of fundamental matrix estimation methods. When the accuracy is not enough, high-cost back-end optimization is required for the correction, and its low efficiency affects the real-time performance of the system. To solve these problems, an improved quasi-affine transformation method is proposed based on the QUATRE algorithm. Firstly, a specific “gene-chromosome” mode is used for the collaboration of the particles. Besides, the way of initialization, mutation, and crossover of the original QUATRE algorithm are redefined within the discrete solution space described by the homogeneous coordinates. In addition, a confidence coefficient based iteration termination method is presented for the acceleration. Experiments show that the proposed method is useful for fundamental matrix estimation. It can effectively get rid of the disturbance of outliers resulting from the noises and mismatches, and it outperforms the methods such as the LMedS, RANSAC, and MSAC in terms of accuracy and efficiency.


      A velocity planning method based on fuzzy-neural network for autonomous driving
      WANG Meng, CHEN Jue-xuan, DENG Zheng-xing
      2021, 43(11): 2011-2019. doi:
      Abstract ( 117 )   PDF (2274KB) ( 164 )     
      In order to improve the comfort performance of autonomous driving and reduce the time complexity of velocity planning algorithm, a longitudinal velocity planning method based on fuzzy neural network is proposed. Manual driving experience is summarized up as a fuzzy rule table, and a fuzzy planning model is established. By utilizing the self-learning function of neural network, the fuzzy planning model is modified, so as to build the fuzzy neural network planning model. Static obstacle scene and dynamic obstacle scene are analyzed. Simulations verify the algorithm feasibility. Compared with the traditional fuzzy planning method, the proposal have smoother acceleration curve. The proposed method has certain anti-disturbance ability, is easy to implement, and ensures the real-time performance and stability.

      Driving conditions and fuel consumption of an improved K-means clustering algorithm
      SU Xiao-hui, ZHANG Yu-xi, XU Shu-ping, SHANG Yu
      2021, 43(11): 2020-2026. doi:
      Abstract ( 111 )   PDF (1181KB) ( 139 )     
      In order to solve the problem that the initial center of traditional clustering algorithm is easy to fall into local optimum and time-consuming. An improved K-means clustering algorithm is proposed. In this algorithm, the maximum minimum distance and weighted Euclidean distance are introduced to avoid the influence of outliers and edge data. The weight method is used to improve the principal component, and the feature influence factor is used as the initial feature weight to construct a weight- ed Euclidean distance measure. According to the influence factors of feature contribution rate on cluster-  ing, a clustering method of feature weight influence factors is proposed, which selects representative feature factors to highlight clustering effect, and finally synthesizes driving cycle and analyzes instantaneous fuel consumption. The results show that: the difference value of speed acceleration joint distribution of the proposed method is only 1.05%, which saves 44.2% of the time, compared with the traditional K-means clustering. The driving cycle fitting degree is high, which can reflect the actual vehicle operation characteristics and fuel consumption.

      Study on the relationship between discrete and continuous enhanced Karnik-Mendel algorithms in type-reduction of higher-order fuzzy system
      CHEN Yang, WANG Tao
      2021, 43(11): 2027-2034. doi:
      Abstract ( 72 )   PDF (858KB) ( 92 )     
      Type-reduction is the key block in general type-2 fuzzy logic systems. This paper compares and analyzes the sum operation in discrete enhanced Karnik-Mendel (EKM) algorithms and the integral operation in continuous enhanced Karnik-Mendel (CEKM) algorithms. Based on the alpha- representation theory of general type-2 fuzzy sets, EKM algorithms are extended to perform the centroid type-reduction of general type-2 fuzzy logic systems. While computing the centroid type-reduced sets and the defuzzified values of general type-2 fuzzy logic systems, two computer simulation examples show that the calculation results of EKM algorithms can accurately approximate the CEKM algorithms, when the number of sampling of primary variables of centroid output GT2 FSs increases appropriately.

      A grey wolf optimization algorithm based on Cubic mapping and its application
      ZHANG Meng-jian, ZHANG Hao, CHEN Xi, YANG Jing
      2021, 43(11): 2035-2042. doi:
      Abstract ( 303 )   PDF (913KB) ( 216 )     
      Aiming at the problem that the grey wolf optimization algorithm (GWO) is easy to fall into the local optimal solution to the complex optimization problems, from the perspective of chaos initia- lization and nonlinear control strategy, a grey wolf optimization algorithm based on cubic mapping and opposition-based learning is proposed (COGWO). Firstly, the cubic mapping and opposition-based learning strategies are used to initialize the population, and the parameters are adjusted by a nonlinear parameter control strategy in the optimization process. Then, the optimization experiment on six benchmark test functions show that the COGWO algorithm has better convergence accuracy, convergence speed and stability. Finally, the COGWO algorithm is applied to a practical engineering optimization problem.
      A deep neural network model compression method based on Adams shortcut connection
      DU Peng, LI Chao, SHI Jian-ping, JIANG Lin
      2021, 43(11): 2043-2048. doi:
      Abstract ( 102 )   PDF (731KB) ( 115 )     
      Deep neural network has made a great breakthrough in all kinds of computer vision tasks, but there is still a lack of guiding principles in network structure design. A great deal of theoretical and empirical evidence shows that the depth of neural networks is the key to their success, but the trainability of neural networks remains to be solved. In this paper, the numerical method of differential equations (Adams) is used in the weight learning of deep neural network, and a shortcut connection based on Adams method is proposed to improve the learning accuracy of late network, compress the size of model, and make the model more effective. In particular, the trainability optimization effect is obvious for the deep neural network with a small number of layers. Taking the classic ResNet as an example, this paper compares the performance between Adams-ResNet, which uses the shortcut connection based on Adams method, and the source model on Cifar10. The former improves the recognition accuracy while reducing the parameters of the source model to half.
      Multi-label feature selection based on label co-occurrence relationship
      LI Yu-chen, WEI Wei, BAI Wei-ming, WANG Da
      2021, 43(11): 2049-2055. doi:
      Abstract ( 147 )   PDF (599KB) ( 141 )     
      Multi-label data widely exists in the real world, and multi-label feature selection is an important preliminary step in multi-label learning. Based on the fuzzy rough set model, researchers have proposed multi-label feature selection algorithms, but most of these algorithms do not pay attention to the co-occurrence characteristics between labels. In order to solve this problem, the similar relationship between the samples under the label set is evaluated based on the co-occurrence relationship between the sample labels. This relationship is used to define the fuzzy mutual information between the feature and the label. Combining the principle of maximum correlation and minimum redundancy, a multi-label feature selection algorithm is designed. Experiments on 5 public data sets show the effectiveness of the proposed algorithm.

      A warning propagation algorithm for solving proposition formula backbones 
      WANG Shuai, WANG Xiao-feng, LIANG Tian, LI Zhi
      2021, 43(11): 2056-2061. doi:
      Abstract ( 98 )   PDF (836KB) ( 116 )     
      The Warning Propagation (WP) algorithm is an important type of message passing algorithm, which is very effective in judging the satisfiability of propositional formulas. Through the analysis of the mathematical principle of the WP algorithm, it is found that when the algorithm converges, the assignment of some arguments is fixed with high probability, so that the formula can be simplified. Based on such features, the iterative equations and argument assignment conditions of WP algorithm are modified, and a message passing algorithm for solving the backbones of propositional formulas is designed. When the number of variables exceeds 400, the proposal improves the efficiency by 40% in comparison to the classic backbone set solving algorithm [18,20,22], and by 10% in comparison to the current commonly used algorithm [19,21]. The results show that this method is very effective in solving the backbones of proposition formulas.