Please wait a minute...

Journal of Beijing University of Posts and Telecommunications

  • EI核心期刊

Current Issue

  • Review

    • The Key Techniques and Future Vision of Feature Selection in Machine Learning
    • CUI Hong-yan, XU Shuai, ZHANG Li-feng, Roy E. Welsch, Berthold K. P. Horn
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 1-12. DOI:10.13190/j.jbupt.2017-150
    • Abstract ( 780 )     HTML       
    • Big data research is widely spread around the world, and feature selection of machine learning plays an important role on these researches. To address the issue of discovering novel feature selection methods in data mining tasks on big data, this paper researches five models related to feature selection:linear coefficient correlation, Lasso sparse selection, ensemble learning models, neural networks, principal component analysis. The merits and drawbacks of these models are extensively discussed in depth in this paper, which may help in providing a direction for those who are interested in the machine learning area.
    • References | Supplementary Material | Related Articles

    Papers

    • Defect Prediction Method for Android Binary Files
    • DONG Feng, LIU Tian-ming, XU Guo-ai, GUO Yan-hui, LI Cheng-ze
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 13-23. DOI:10.13190/j.jbupt.2017-243
    • Abstract ( 462 )     HTML       
    • Software defect prediction is an important method in the field of software security. Most of existing defect prediction models are source-oriented and can not be easily used for Android binary files (apks) defect prediction. Moreover, the traditional machine learning techniques used in these models have a shallow architecture, which leads to a limited capacity of expressing complex functions between features and defects. The author proposes a practical defect prediction model for Android binary files using deep neural network (DNN). A new approach is proposed to generate features that capture both token and semantic features of the defective smali (decompiled files of apks) files in apks. The feature vectors are input into DNN to train and build the defect prediction model in order to achieve accuracy. The article implements the model called DefectDroid and applies it to a large number of Android smali files. The performance of DefectDroid is compared from three aspects:within-project defect prediction, cross-project defect prediction and traditional machine learning algorithms.
    • References | Supplementary Material | Related Articles
    • Selection of Hybrid Color Space for Skin Detection Based on Feature Selection Method
    • LIU Xin-hua, ZHAO Zi-qian, KUANG Hai-lan, MA Xiao-lin, LI Fang-min
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 24-30. DOI:10.13190/j.jbupt.2017-184
    • Abstract ( 340 )     HTML       
    • To solve the problem of selecting appropriate color space for skin detection, a feature-selection-based method is exploited and two improvements on traditional method are proposed:Firstly, mutual information is used to narrow the feature selection range, then feature subset which produces the best classification accuracy will be selected; Secondly, a variety of possible feature subset initialization schemes are tested, and then choose feature set that reveal the best result. Experimental results and comparative analysis show that the hybrid color spaces obtained by the improved feature selection method have better skin detection performance than traditional color spaces and existing hybrid color spaces.
    • References | Supplementary Material | Related Articles
    • Uniqueness and Theory Bounds of Two-Dimensional Correlation Frequency Hopping Sequence Pair
    • XU Cheng-qian, ZHAO Ya-jie
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 31-36. DOI:10.13190/j.jbupt.2017-097
    • Abstract ( 299 )     HTML       
    • In view of the traditional frequency hopping sequence pair containing only one-dimensional Hamming correlation function of time-delayed variables, frequency shift factors are added to correlation function,and the concept of time-frequency two-dimensional correlation function of frequency hopping sequence pair is proposed. The uniqueness of frequency hopping sequence pair is proved, and the only reception of this kind of signal is guaranteed in the practical application. The theory bounds formed by two-dimensional Hamming correlation values, the number of frequency hopping sequence pair, the number of frequency gap and sequence length are derived. It is of great significance to construct the two-dimensional correlation frequency hopping sequence pair which can satisfy the theory bounds.
    • References | Supplementary Material | Related Articles
    • Point-of-Interest Recommendation with Spatio-Temporal Context Awareness
    • XU Qian-fang, WANG Jia-chun, XIAO Bo
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 37-42,50. DOI:10.13190/j.jbupt.2017-081
    • Abstract ( 424 )     HTML       
    • A personalized hybrid point-of-interest recommendation with spatio-temporal context awareness was proposed to provide users in location-based social networks with superior service. Spatially, two-dimension kernel density estimation was performed for each cluster of check-ins derived by hierarchical clustering and averaged. Meanwhile, random walk on graph was iterated on transition matrices generated from sequence information, location information and social network. The hybrid model combines spatio-temporal context above for recommendation. Experiment on real-world location-based social network(LBSN) datasets demonstrates that the performance metrics of precision and recall of the hybrid recommendation model is superior to other baseline techniques in both standard recommendation scene and cold-start scene.
    • References | Supplementary Material | Related Articles
    • Lyapunov-Based Virtual Resource Allocation in Wireless Networks with Self-Backhauls
    • TANG Lun, YANG Xi-xi, SHI Ying-jie, CHEN Qian-bin
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 43-50. DOI:10.13190/j.jbupt.2017-165
    • Abstract ( 358 )     HTML       
    • To improve the flexibility of network deployment and satisfy the diversity of virtual network demand, a virtual resource allocation strategy utilizing Lyapunov for wireless self-backhaul network was proposed. Firstly, a joint radio access resources and backhaul algorithm is deployed to maximize the virtual network utility under certain practical preconditions, i. e. the network queue stability, the minimum average data rate for each virtual network and the capacity constraint of the backhaul link. Secondly, a real-time scheduling algorithm based on the current channel state and queue state is designed by Lyapunov optimization, Lagrange duality algorithm and particle swarm algorithm based on similarity random variation. Simulation shows that the proposed method can effectively improve the total average revenue of virtual network while guaranteeing the queue stability.
    • References | Supplementary Material | Related Articles
    • Method of Diamond Supplement for Indoor Location Micro Base Station Placement
    • WANG Hui-qiang, LIU Xiu-bing, LÜ Hong-wu, FENG Guang-sheng, YANG Yan-ping
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 51-58,87. DOI:10.13190/j.jbupt.2017-175
    • Abstract ( 368 )     HTML       
    • Due to non-line-of-sight and multipath effects of the indoor scene,the target area to be positioned after the placement of the outdoor base station will cover the loopholes,resulting in reduced positioning accuracy. The article proposed an algorithm based on diamond-shaped layout to compensate the loopholes in the target area. Firstly,the base station placement algorithm applied to the outdoor base station placement was thought the combination of genetic algorithm and ant colony algorithm. In the second place,the target area of the signal coverage and add micro-network element was analyzed to cover the loopholes. Experiment shows that the 90% average localization error of the target area is within 2m,which is 10% higher than the existing immune algorithm.
    • References | Supplementary Material | Related Articles
    • A Novel Method for Depicting Connectivity in Opportunistic Sensor Networks
    • SHU Jian, JIANG Shan-dong, SUN Li-min
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 59-64. DOI:10.13190/j.jbupt.2017-148
    • Abstract ( 229 )     HTML       
    • Connectivity is an important metric to reflect the performance of network. Connectivity of opportunistic sensor networks (OSN) is of temporal evolution, which is hard to be modeled with traditional graph model. Connectivity of OSN is modeled with temporal graph, and network connectivity degree is achieved by computing temporal path, temporal distance, and connectivity effectiveness. Furthermore, network connectivity degree is proposed to depict connectivity of OSN. The simulation results show that network connectivity degree can reflect connectivity of OSN more accurately in different experiment scenarios.
    • References | Supplementary Material | Related Articles
    • An Entity Discover and Linking Approach Based on Convolutional Neural Network and Random Walk with Restart
    • TAN Yong-mei, LI Xiao-guang, LÜ Xue-qiang
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 65-69. DOI:10.13190/j.jbupt.2017-127
    • Abstract ( 443 )     HTML       
    • An entity linking approach based on convolutional neural network and random walk with restart was presented. This method first discovers the mentions in the text, after generates the mention candidate entity set, then selects the candidate entity using the entity linking approach based on convolutional neural network and random walk with restart and clusters the mentions those do not have the corresponding entity in the knowledge base. Our method FCEAFm is 0.652 on the TAC-KBP2016 entity discovery and linking evaluation data set, and the first team is 0.643. The results show our method can effectively solve this problem.
    • References | Supplementary Material | Related Articles
    • TOChain: a High-Performance SFC for Virtual Network Security
    • TANG Hong-wei, FENG Sheng-zhong, ZHAO Xiao-fang
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 70-80. DOI:10.13190/j.jbupt.2017-134
    • Abstract ( 477 )     HTML       
    • Performance problem is a big challenge for network function virtualization (NFV) based security service function chain (NS-SFC). To solve this problem, a TCP offloading based SFC for virtual network security, called TOChain was proposed, which avoids reduplicative packet processing over TCP/IP stack and virtual network interfaces. And furthermore, a throughput guarantee oriented strongly synchronized periodical CPU scheduling algorithm for TOChain was presented. Finally, the prototype based on KVM and the performance of the prototype with three types of virtualized network function (VNF), including iptables, Snort and FreeWAF was developed and evaluated. It is shown that TOChain achieves a significantly higher performance with a lower CPU utilization compared to the NFV based traditional SFC architecture. With strongly synchronized periodical algorithm, the network performance achieved is very close to the configured throughput under the light and medium traffic load. Moreover, even under the heavy load, it also ensure fairness between virtual machines.
    • References | Supplementary Material | Related Articles
    • L1-Nonlocal Means Regularization Model for Image Deblurring Problem
    • FENG Xiang-chu, LIU Xin, YANG Chun-yu, WANG Wei-wei
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 81-87. DOI:10.13190/j.jbupt.2017-101
    • Abstract ( 390 )     HTML       
    • An l1-nonlocal means regularization model was proposed in order to preserve the edges and details while deblurring the blurred image. Firstly, the article empirically gave out that the distribution of the residual in the nonlocal means denoising algorithm (differences between the noisy image and the denoised result) is heavy-tailed, which well fits the Laplacian distribution. Based on this observation, a new regularization model was proposed by using the l1-norm constrained residual as the new regularization term. Then the corresponding optimization algorithm was designed by utilizing the Bregmanized operator splitting algorithm, which can be regarded as an extension of plug-and-play Priors algorithm. Experiments show that the new model achieves better performance than the l2-nonlocal means regularization model and the plug-and-play priors model in terms of both restoration results and preserving the edges and details of the image.
    • References | Supplementary Material | Related Articles
    • Model Construction of Uygur and Korean Morphological Analysis
    • XU Chun, JIANG Tong-hai, YU Kai, JIANG Wen-bin
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 88-94. DOI:10.13190/j.jbupt.2017-117
    • Abstract ( 548 )     HTML       
    • A discriminant model of the graphic structure is established for the morphological analysis of Uighur and Korean language. The model builds the morphological analysis of the sentence into the graphic structure of morphological components, and describes the correlation between the morphological components of the words inside and the morphological components of the adjacent words through flexible and rich feature design. Compared with the traditional linear model, the pattern model is better to consider the linguistic association between the morphological components, and it is expected to achieve higher sentence analysis performance. The experimental results in Uighur and Korean indicate that the graphic model achieves certain performance improvement comparing with the linear model, and the word level accuracy of morphological analysis respectively increases by 4.4 and 2.8 percentage points.
    • References | Supplementary Material | Related Articles

    Reports

    • Evidence Combination Method Based on Singular Value and Falsity
    • XUE Da-wei, WANG Yong, GAO Kang-kai
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 95-102. DOI:10.13190/j.jbupt.2017-057
    • Abstract ( 311 )     HTML       
    • Most researchers think that modifying evidence source based methods are more reasonable to deal with the problem of conflicting evidence combination. However, most existing methods of modifying evidence source, which generally evaluate the evidence from single angle, have some deficiencies. To resolve such a problem, a new evidence combination method based on singular value and falsity is proposed. Firstly, conflict between two evidences is measured by the minimum singular value of the matrix composed of basic probability assignment (BPA) corresponding to two evidences, on the basis of which the definition of credibility of evidence is presented. Then, the weights which are used to average BPAs of evidences are produced by modifying the credibility with the falsity of evidences. Finally, the weighted average of BPAs is fused by Dempster's rule of combination. The numerical examples illustrate that the presented method can combine conflicting evidences effectively, and has faster convergence speed and better focusing degree than some existing methods.
    • References | Supplementary Material | Related Articles
    • Application of Curved Surface Ray Tube in the Prediction of Complex Electromagnetic Environment
    • LÜ Na, SHI Dan, JIANG Wei, GAO You-gang, TANG Chao-han
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 103-108. DOI:10.13190/j.jbupt.2017-142
    • Abstract ( 447 )     HTML       
    • A ray tracing method for predicting radio propagation based on a new curved surface ray tube was presented. Compared with the existing quadrilateral wave front diffraction ray tube, the curved surface ray tube can effectively reduce the calculation time and improve the efficiency during the tracing process. Calculation verified the high efficiency of the curved surface ray tube model, a 4 times speed up compared to the four-ray tube model was achieved. The method to deal with the intersection of the ray tube and the concave terrain was also introduced. Moreover, the wave propagations in several different environments were simulated with our developed software based on the curved surface ray tube and four-ray tube tracing method. Compared with the four-ray tube model, the average computational time of the cured surface ray tube model is shortened by about 18.49%, and the average error is only 2.93%. The simulation results prove the improvement of the efficiency and the accuracy of the tracing process.
    • References | Supplementary Material | Related Articles
    • Object-Oriented Software Coupling Metrics
    • MA Jian, LIU Feng, FAN Jian-ping
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 109-114. DOI:10.13190/j.jbupt.2017-143
    • Abstract ( 471 )     HTML       
    • The article analyzes the shortage of C&K software metrics suit for an improvement, then proposes a decomposed coupling metrics for objected-oriented (DCMOO). The metrics refer to unified modeling language class diagram to analyze the relationships between classes in software design, and evaluate the quality of the software by using a set of formal evaluation theorems, It is shown that DCMOO can satisfy these theorems. Finally the article uses JUnit and JEdit as research object, and applies DependencyFinder and Eclipse Metrics to calculate the proposed software coupling metrics automatically meanwhile validate the proposed metrics.
    • References | Supplementary Material | Related Articles
    • Neural Segmentation Method of Ultrasound Image
    • XU Chen-yang, LI Meng-xin, YANG Juan
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 115-120. DOI:10.13190/j.jbupt.2017-151
    • Abstract ( 342 )     HTML       
    • To improve the efficiency of neural segmentation in ultrasound images, we propose a new neural structure the U-shape residual network. Compared with U-net network, this structure deepens the original structure to improve the expression ability. By standardizing the parameters of each layer, the structure reduces the training time and improve the segmentation effect. According to the results, the U-shape residual network segmentation effect increased by 13% compared with U-net network and improved about 7% compared with SegNet network.
    • References | Supplementary Material | Related Articles
    • Stability Analysis for a Modified Bilinear Model Based on Recursive Least Squares
    • ZHAO Xia, NI Ying-ting, LI Zhan-ning
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 121-124. DOI:10.13190/j.jbupt.2016-272
    • Abstract ( 311 )     HTML       
    • The nonlinearity of bilinear polynomial only relies on the input-output cross term to express; it is hard to accurately describe the system with higher order nonlinearity. For improving the performance of models, many modified bilinear polynomials are proposed. However, the complex feedback terms cause models to be unstable, which restrict the modified models to be widely used in practice. Therefore, the stability of a modified bilinear model was analyzed. Recursive least squares (RLS) is used to identify parameters of the modified bilinear model, and the iterative formulas of the algorithm are deduced in the complex field. Simultaneously, the stability of the identified system is verified. It is shown that the bilinear system model identified by RLS, has bounded-input bounded-output stability.
    • References | Supplementary Material | Related Articles
    • Improved Multi-Identity Based Fully Homomorphic Encryption Scheme over Lattices
    • TANG Yong-li, HU Ming-xing, YE Qing, QIN Pan-ke, YU Jin-xia
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 125-133. DOI:10.13190/j.jbupt.2017-163
    • Abstract ( 409 )     HTML       
    • Aiming at low efficiency of trapdoor function in multi-identity based fully homomorphic encryption (mIBFHE) schemes, a new mIBFHE scheme was proposed. Firstly, the MP12 trapdoor function with Dual-Regev algorithm was combined to construct a transformable identity-based encryption (IBE) scheme, and a Mask system which supports to transform IBE scheme presented to mIBFHE scheme under standard model. Then, based on presented Mask system and eigenvector idea, the IBE schemes was transformed to mIBFHE scheme. Comparing with the similar schemes, the efficiency of the scheme is improved in trapdoor generation and preimage sampling stage, and the lattice dimension, the size of ciphertext and evaluated ciphertext, etc. are obviously reduced. The security of the presented scheme strictly is reduced to the hardness of learning with errors problem in the standard model.
    • References | Supplementary Material | Related Articles
    • A Link Quality Estimation Method for WSNs Based on Extreme Learning Machine
    • LIU Lin-lan, XU Jiang-bo, CHEN Yu-bin, SHU Jian
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 134-138. DOI:10.13190/j.jbupt.2017-185
    • Abstract ( 304 )     HTML       
    • An approach of estimating link quality was proposed which is based on extreme learning machine. The index of link asymmetry, the coefficient of variation of signal to noise ratio and mean signal to noise ratio are chosen as link quality parameters. Link quality level is classified by link packet receive rate which is the evaluation index. Particle swarm optimization algorithm is employed to optimize input weights and offset parameter, so that link quality model is built. In different scenarios, compared with the support vector classification machine estimate methods, the experimental results show that the proposed estimation method achieves better precision.
    • References | Supplementary Material | Related Articles
    • Joint Power Control and Beamforming Optimization for MIMO-OFDM Network
    • HUANG Miao-na, CHEN Jun, REN Bin
    • Journal of Beijing University of Posts and Telecommunications. 2018, 41(1): 139-144. DOI:10.13190/j.jbupt.2017-026
    • Abstract ( 512 )     HTML       
    • A joint optimization problem of power control and beamforming for multiple input multiple output (MIMO)-orthogonal frequency division multiplexing (OFDM) system with more realistic interference model was proposed. Then a two-stage scheme based on dual of uplink-downlink and standard interference mapping theory was given to solve this problem, which is proved to be rigorously convergent. Simulations show that the proposed scheme converges faster in multiple-antenna cases and the proposed algorithm outperforms the traditional ones in terms of power saving.
    • References | Supplementary Material | Related Articles