Home      About Journal      Editorial Board      Instructions      Ethics Statement      Contact Us                   中文

Judgement and Disposal of Academic Misconduct Article Copyright Transfer Agreement Proof of Confidentiality Requirements for Electronic Version More.... Chinese Association of Automation National ResearchCenter for Intelligent Computing System Institute of Intelligent Machines,Chinese Academy of Sciences More....

2016 Vol.29 Issue.7, Published 2016-07-30 Papers and Reports    Researches and Applications        Papers and Reports 577 Stochastic Algorithm with Reduced Variance and Weighted Average for Solving Non-smooth Strongly Convex Optimization Problems ZHU Xiaohui, TAO Qing Using the strategy of reducing the variance in smooth stochastic method can effectively improve the convergence of the algorithm. An algorithm, hybrid regularized mirror descent with reduced variance and weighted average (α-HRMDVR-W), is obtained by using weighted average and reduced variance for solving “L1+ L2 + Hinge” non-smooth strong convex optimization problem. The variance reduction strategies are utilized at each step of the iterative process, and the weighted average of the output mode is used. It is proved that the α-HRMDVR-W has optimal convergence rate and the convergence rate does not depend on the number of samples. Unlike the existing variance reduction methods, α-HRMDVR-W only uses a small portion of samples instead of the total samples to modify the gradient at each iteration. Experimental results show that α-HRMDVR-W reduces the variance and decreases CPU time. 2016 Vol. 29 (7): 577-589 [Abstract] ( 671 ) [HTML 1KB] [ PDF 556KB] ( 896 ) 590 Feature Selection Algorithm Based onGranulation-Fusion for Massive High-Dimension Data JI Suqin, SHI Hongbo, Lü Yali, GUO Min From a granular computing perspective, a feature selection algorithm based on granulation-fusion for massive and high-dimension data is proposed. By applying bag of little Bootstrap (BLB), the original massive dataset is granulated into small subsets of data (granularity), and then features are selected by constructing multiple least absolute shrinkage and selection operator(LASSO) models on each granularity. Finally, features selected on each granularity are fused with different weights, and feature selection results are obtained on original dataset through ordering. Experimental results on artificial datasets and real datasets show that the proposed algorithm is feasible and effective for massive high-dimension datasets. 2016 Vol. 29 (7): 590-597 [Abstract] ( 568 ) [HTML 1KB] [ PDF 421KB] ( 574 ) 598 Support Tensor Machine Classifier with Pinball Loss YU Keming, HAN Le, YANG Xiaowei The input patterns are usually high-order tensors in the fields of machine learning, pattern recognition, data mining, etc. In this paper, the pin-support vector machine is firstly extended from vector to tensor and the support tensor machine (STM) classifier with pinball loss(pin-STM) is proposed. Then, a sequential minimal optimization (SMO) algorithm is designed to solve this model. To maintain the nature structure of tensor and speed up the training procedure, the rank-one decomposition of tensor is used to substitute the original tensor to compute the inner products of tensors. The experimental results on vector datasets and tensor datasets show that SMO is faster than the classical active-set method for vector data. Compared with pin-SVM, the pin-STM has higher training speed and better generalized performance for tensor data. 2016 Vol. 29 (7): 598-607 [Abstract] ( 763 ) [HTML 1KB] [ PDF 475KB] ( 761 ) 608 Nonnegative Matrix Factorization Algorithm with Prior Information for Community Detection LI Guopeng, PAN Zhisong, YAO Qing, LI Deyi To solve the problem of community detection in complex networks, a semi-supervised nonnegative matrix factorization (NMF) algorithm with prior information is proposed to obtain more accurate and better understanding results, and the detailed iteration algorithm is presented. In this algorithm, prior information is added to object function as additional constraints in community indicator matrix. Consequently, results are more meaningful. The experiments on real-world network datasets confirm the effectiveness of the proposed algorithm. It reduces the negative impact of the addition of prior information on node importance analysis with NMF, and it is suitable for weighted and un-weighted networks. 2016 Vol. 29 (7): 608-615 [Abstract] ( 533 ) [HTML 1KB] [ PDF 519KB] ( 788 ) 616 Vector Autoregression Model Based Microblog Hidden Topic Popularity Prediction DUAN Dongsheng, LI Pengxiao, LI Yuhua, LI Ruixuan The existing topic popularity prediction methods predict the topic popularity just based on the features of topic and the correlations between different topics are not taken into account. However, there are correlations among different topics in microblog contexts, especially for the topics of the same event. Aiming at this problem, dynamic topic model is firstly employed to detect the hidden topics and their popularity time series from microblogs in this paper. Then, the Jensen-Shannon divergence and Pearson′s correlation coefficient are computed to investigate the correlations among topic contents and among topic time-series, respectively. Thus, the motivation of introducing topics correlation is revealed. Finally, a vector auto-regressive (VAR) model based Microblog hidden topic popularity prediction algorithm is proposed by introducing correlations among different topic time-series in model training. Experiments are conducted on the real data. Experimental results show that the proposed algorithm performs better in prediction accuracy and model fitting than algorithms without consideration of correlations among different topics. 2016 Vol. 29 (7): 616-624 [Abstract] ( 402 ) [HTML 1KB] [ PDF 891KB] ( 595 ) Researches and Applications 625 Shifted Label Proportion Aware Semi-supervised Support Vector Machine LI Yuanzhao, WANG Shaobo, LI Yufeng When the label proportion of unlabeled data is far away from that of labeled data, direct supervised support vector machine(SVM) with only labeled data outperforms semi-supervised SVM(S3VM) with unlabeled data. Thus, a shifted label proportion aware S3VM(fairS3VM) is proposed. Specifically, the label mean of unlabeled data is firstly estimated. Then multiple label means corresponding to multiple label proportions are integrated under the worst-case scenario. Experimental results show that the performance guarantee of S3VMs is effectively improved when the label proportion is shifted. 2016 Vol. 29 (7): 625-632 [Abstract] ( 442 ) [HTML 1KB] [ PDF 386KB] ( 514 ) 633 Least Squares Semi-supervised Support Tensor Machine LU Chengtao, LI Fanzhang, ZHANG Li, ZHANG Zhao Support tensor machine has a high computational complexity due to the iterative procedure. To overcome the shortcoming, the optimization is modified , the model is trained by solving a set of linear equations instead of solving a quadratic program problem. Additionally, transductive method is used to solve the semi-supervised problem, least squares semi-supervised support tensor machine is proposed. Some experiments on face recognition and time series classification are conducted to compare the proposed algorithm with the traditional algorithms. The results show that the proposed algorithm reduces the computation time and improves the recognition rate. 2016 Vol. 29 (7): 633-640 [Abstract] ( 739 ) [HTML 1KB] [ PDF 466KB] ( 871 ) 641 Feature Selection Framework of Whole-Brain Functional Magnetic Resonance Imaging Data Based on Regularized Softmax Regression QU Yongkang, JI Junzhong, LIANG Peipeng, GAO Mingxia To solve the classification model overfitting problem caused by the high dimension and small sample properties of functional magnetic resonance imaging (fMRI) data, a feature selection framework of whole-brain fMRI data combining L1-norm regularization and L2-norm regularization in softmax regression is proposed. Firstly, the whole brain is divided into the region of interest (ROI) and the region of non-interest (RONI) in terms of the characteristics of brain cognition. Then, L2-norm regularization shrinking the weighting coefficients is used to model all voxels in ROI while L1-norm regularization with a sparse effect is employed for modeling the activated voxels in RONI. Finally, the regularized softmax regression model of whole-brain fMRI data is constructed by integrating all voxels in ROI and the activated voxels in RONI. The experimental results on Haxby datasets show that the regularization strategies of L2-norm and L1-norm effectively improve the whole-brain classification performance compared to some other methods. 2016 Vol. 29 (7): 641-649 [Abstract] ( 417 ) [HTML 1KB] [ PDF 1272KB] ( 668 ) 650 Discrete Mussels Wandering Optimization Algorithm for Solving Traveling Salesman Problem HAN Wei, ZHANG Zicheng A discrete mussels wandering optimization (DMWO) algorithm is designed for solving the traveling salesman problem (TSP). An evaluation function and a measure operator indicating differences among mussels are given within DWMO framework. To overcome the defect caused by the overall discrete routine adjustment, a local routine adjustment strategy based on 2-opt is adopted to enhance the searching ability of the algorithm. The experiment is conducted on several standard TSPLIB testing data of different sizes. Compared with discrete glowworm swarm optimization and ant colony optimization adopting 2-opt, the results show the competitive performance of the proposed algorithm in terms of solution consistency, accuracy and the number of iterations. 2016 Vol. 29 (7): 650-657 [Abstract] ( 496 ) [HTML 1KB] [ PDF 444KB] ( 523 ) 658 Effective Distance Based Multi-modality Feature Selection YE Tingting, LIU Mingxia, ZHANG Daoqiang Based on the traditional distance measurements, effective distance is adopted to implement feature selection for multi-modality classification. To better reflect the global and local relationships among samples, an effective distance based multi-modality feature selection method is proposed. This method focuses on the global relationship among samples to build model, and effective distance based feature selection learning is realized. Thus, discriminative features are selected. To evaluate the efficiency of the proposed method, experiments are performed on the Alzheimer's disease neuroimaging initiative database and the UCI benchmark database. The experimental results demonstrate that compared with traditional feature selection methods using the Euclidean distance, the proposed method significantly improves the results of multi-modality classification. 2016 Vol. 29 (7): 658-664 [Abstract] ( 552 ) [HTML 1KB] [ PDF 595KB] ( 696 ) 665 Classification Algorithm Combined with Unsupervised Learning for Data Stream XU Shuliang , WANG Junhong An ensemble learning techniques based algorithm combined with unsupervised learning is proposed for concept drift problem of data stream. An attribute reduction mechanism is introduced into classification process and then a clustering algorithm is applied to the data for clustering. Accuracies of classification and clustering are compared to decide whether concept drift appears or not. The experimental results show that the proposed algorithm efficiently decreases time consumption and improves the precision. 2016 Vol. 29 (7): 665-672 [Abstract] ( 448 ) [HTML 1KB] [ PDF 438KB] ( 749 )

Supervised by China Association for Science and Technology Sponsored by Chinese Association of Automation NationalResearchCenter for Intelligent Computing System Institute of Intelligent Machines, Chinese Academy of Sciences Published by Science Press

Copyright © 2010 Editorial Office of Pattern Recognition and Artificial Intelligence Address: No.350 Shushanhu Road, Hefei, Anhui Province, P.R. China Tel: 0551-65591176 Fax:0551-65591176 Email: bjb@iim.ac.cn Supported by Beijing Magtech 　Email:support@magtech.com.cn