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....

2017 Vol.30 Issue.9, Published 2017-09-30 Orignal Article        Orignal Article 769 Attribute Reduction of Formal Contexts Based on Decision Rules LI Tongjun, XU Yingcong, WU Weizhi, GU Shenming By using the binary relation between objects and attributes, one pair of lower and upper rough fuzzy approximation operators is defined in formal contexts, properties of the approximation operators are explored, and the relationship between the defined approximation operators and the existing rough approximation operators is revealed. By using the defined approximation operators, two types of decision rules can be extracted, i.e., the certainty rules and the possibility rules. Subsequently, with respect to two types of decision rules, notions of lower and upper approximation reductions are proposed. For the upper approximation reduction, some necessary and sufficient conditions for reducible attributes and consistent subsets of attributes are obtained. An approach for attribute reduction is presented, and some illustration examples are given to show its reliability. 2017 Vol. 30 (9): 769-778 [Abstract] ( 603 ) [HTML 1KB] [ PDF 563KB] ( 352 ) 779 Elastic Kernel Subspace Clustering ZHANG Pengtao, CHEN Xiaoyun In the existing subspace clustering algorithms, it is assumed that the data is derived from a union of multiple linear subspace, and these algorithms cannot deal with problems of nonlinear and time warping in time series clustering. To overcome these issues, elastic kernel low rank representation subspace clustering(EKLRR) and elastic kernel least squares regression subspace clustering(EKLSR) are proposed by introducing kernel tricks and elastic distance, and they are called elastic kernel subspace clustering(EKSC). Moreover, the grouping effect of EKLSR and the convergence of EKLRR are proved theoretically. The experimental results on five UCR datasets show the effectiveness of the proposed algorithms. 2017 Vol. 30 (9): 779-790 [Abstract] ( 459 ) [HTML 1KB] [ PDF 729KB] ( 440 ) 791 Optimal Matching Tracking Algorithm Based on Discriminant Appearance Model LIU Wanjun , LIU Daqian, FEI Bowen Traditional model matching and tracking algorithms are easily influenced by the occlusion of other targets and the complex background. To solve these problems, an optimal matching tracking algorithm based on discriminant appearance model is proposed. Firstly, the local feature blocks of the previous 5 frames of the image sequences are extracted by sampling, and the training sample set consisting of a number of feature blocks is established. Then, the feature blocks with the same color and texture features are clustered to build a discriminant appearance model. Secondly, the bi-directional optimal similarity matching method is adopted for target detection. To avoid complex background interference, a method of foreground partition is proposed to acquire more accurate matching results. Finally, the tracking results are periodically added to the clustering collection to update the appearance model. The experimental results indicate that the proposed approach provides higher tracking accuracy under the conditions of partial occlusion and complex background by using the discriminant appearance model of multi-frame training and the bi-directional optimal similarity matching method. 2017 Vol. 30 (9): 791-802 [Abstract] ( 453 ) [HTML 1KB] [ PDF 3443KB] ( 369 ) 803 Graph Mining Model Using Markov Clustering Based on Annular Network Motifs REN Yonggong, SUO Quanming, LIU Yang Aiming at the problems of low efficiency and accuracy of data mining, a graph mining model using Markov clustering based on annular network motifs is proposed. Firstly, the Erdo″s-Rényi model is employed to generate random graphs according to the vertices set of the input graph. Annular sub-graphs are judged by the additive property of vectors in the process of sub-graph mining from input and random graphs. In the next step, the motif statistical characteristics are calculated and used to label the annular motifs. Then, the correlation matrix of absolute contribution of edges is solved in the graph, and the threshold is obtained by dynamic threshold method for the binarization of matrix. Finally, inflation and expansion processes are carried out on the sparsified graph data to achieve the state of convergence. Experimental results show that the proposed model can effectively reduce the running time and improve the mining efficiency of the graph with the guaranteed clustering quality. 2017 Vol. 30 (9): 803-814 [Abstract] ( 388 ) [HTML 1KB] [ PDF 888KB] ( 332 ) 815 Linearity Property Testing Approach to Gaussian Kernel Selection HAN Zhizhuo, LIAO Shizhong Kernel selection is critical to the performance of kernel methods. The computational complexity of the existing approaches to Gaussian kernel selection is Ω(n2). Therefore, it is an impediment to the development of large-scale kernel methods. To address this issue, a linearity property testing approach to Gaussian kernel selection is proposed. Completely different from the existing approaches, the proposed approach only needs O(ln(1/δ)/ 2) query complexity, and its computational complexity is independent of the sample size. Firstly, a concept called linearity level is defined. It is proved that linearity level can approximate the distance between a function and the linear function class, and the linearity property testing criterion for Gaussian kernel selection is presented via the concept of linearity level and the approximate distance. The linearity property testing criterion can be applied in random Fourier feature space to assess and select a suitable Gaussian kernel. Theoretical and experimental results demonstrate that the linearity property testing approach to Gaussian kernel selection is feasible and effective. 2017 Vol. 30 (9): 815-821 [Abstract] ( 426 ) [HTML 1KB] [ PDF 587KB] ( 415 ) 822 Fast Feature Selection for Functional Data MA Chen, WANG Wenjian, JIANG Gaoxia Feature selection of functional data aims to choose those features slightly correlated and strongly representative, from the huge functional information. And it can simplify the calculation and improve the generalization ability. Traditional feature selection methods are directly applied in functional data, and the results are not effective or efficient. A functional data oriented fast feature selection(FFS) method integrating principal component analysis(PCA) and minimum convex hull is proposed in this paper. FFS can obtain stable subset of features fleetly. Considering the correlation embedding in features, the result of FFS can serve as initial feature subset of other iterative approaches. This means twice feature selection will be needed. As a popular feature selection method for functional data, conditional mutual information(CMI) is adopted. The experiment results on UCR datasets demonstrate the effectiveness of FFS, and a selection strategy under different demands of time cost or classification accuracy is given through the contrast experiments. 2017 Vol. 30 (9): 822-832 [Abstract] ( 490 ) [HTML 1KB] [ PDF 974KB] ( 312 ) 833 Multi-label Learning Model Based on Multi-label Radial Basis Function Neural Network and Regularized Extreme Learning Machine SHAN Dong, XU Xinzheng Extreme learning machine(ELM) possesses the characteristics of fast training and good generalization ability compared with radial basis function neural network(RBFNN), and the affinity propagation(AP) clustering algorithm can automatically determine the number of clusters without a prior knowledge. Therefore, a multi-label learning model named ML-AP-RBF-RELM is proposed, and AP clustering algorithm, multi-label back propagation neural network (ML-RBF) and regularized ELM (RELM) are integrated in this model. ML-RBF is used to map in the input layer. In the hidden layer, the number of hidden nodes can be automatically determined by the sum of clustering centers of AP algorithm, and the center of the RBF function can be computed through the center of K-means clustering algorithm while the clustering number is determined by AP algorithm. Finally, the weights from hidden layer to output layer are rapidly calculated through RELM. The simulation results demonstrate that ML-AP-RBF-RELM performs well. 2017 Vol. 30 (9): 833-840 [Abstract] ( 544 ) [HTML 1KB] [ PDF 607KB] ( 301 ) 841 Hyper-Network Guided Correlation Analysis on Imaging Genetics LI Chanxiu, HAO Xiaoke, ZHANG Daoqiang Imaging genetic studies focus on feature extraction from brain regions-of-interest, but few of them successfully depict the relations among brain areas. It is studied recently that brain properties can be better reflected by adopting a structured network model to quantify the complex connection among brain areas. Accordingly, a hyper-network guided sparse multi-task canonical correlation analysis algorithm is proposed in this paper. Firstly, a sparse representation method is employed and the resting-state fMRI time series are used to construct the hyper-network. Then, three clustering coefficients are extracted from the hyper-network as brain imaging characteristics. Finally, the sparse multi-task canonical correlation analysis is used to acquire the link between genes and three types of image features. The experimental results on ADNI dataset show that the proposed algorithm is helpful to improve the performance of analyzing associations between genotype and phenotype data and discovering some genetic risk factors closely related to the disease. 2017 Vol. 30 (9): 841-849 [Abstract] ( 714 ) [HTML 1KB] [ PDF 1013KB] ( 591 ) 850 Finger Vein Recognition Based on Best Local Difference Code Bit XI Xiaoming, YIN Yilong, ZHANG Mengyu, YANG Lu, MENG Xianjing, DU Hengfang In finger vein recognition, the local details of the existing code based features are ignored and the discrimination information cannot be fully used. To solve these problems, a best local difference code bit(BLDCB) method is proposed for finger vein recognition. The local difference code(LDC) extraction method is developed to extract the codes. Then, the best bit mining criterion based on relation and inter-class divergence is designed for mining best bits in extracted codes. The conditional probability is calculated for capturing relation between bits and subjects and mining robust bits. Consequently, the intra-class divergence is used for mining discriminative bits from the robust bits, and the selected bits are used as the best bits. Therefore, the best bits are more robust and discriminative. The experimental results on PloyU database and self-constructed finger vein database demonstrate the effectiveness and efficiency of the proposed method. 2017 Vol. 30 (9): 850-858 [Abstract] ( 597 ) [HTML 1KB] [ PDF 1466KB] ( 435 ) 859 Three-Way Decisions Model for Multi-object Optimization Based on Confusion Matrix XU Jianfeng, MIAO Duoqian, ZHANG Yuanjian In consideration of the generalized application of confusion matrix as an important algorithmic measurement tool in machine learning field, a three-way decision measure system of the probabilistic rough set is constructed based on three-way decision confusion matrix. Then, the properties of partial three-way decision measures are discussed. A multi-object optimization function model for three-way decisions thresholds computing is proposed as well. In this model, multi-object optimization functions are considered as weighted sums of three-way decisions measures ,and a new semantic interpretation is acquired for solving the optimal threshold. Finally, the solving process of accepting and rejecting thresholds of the model is demonstrated via an case. By comparing with the classic Pawlak rough set method and confusion matrix model, the confusion matrix model can better balance the accurate rate and the commitment rate for three-way decisions. 2017 Vol. 30 (9): 859-864 [Abstract] ( 532 ) [HTML 1KB] [ PDF 512KB] ( 313 )

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