模式识别与人工智能
Home      About Journal      Editorial Board      Instructions      Ethics Statement      Contact Us                   中文
Pattern Recognition and Artificial Intelligence
22 Judgement and Disposal of Academic Misconduct Article
22 Copyright Transfer Agreement
22 Proof of Confidentiality
22 Requirements for Electronic Version
More....
22 Chinese Association of Automation
22 National ResearchCenter for Intelligent Computing System
22 Institute of Intelligent Machines,Chinese Academy of Sciences
More....
 
 
2019 Vol.32 Issue.11, Published 2019-11-25

Papers and Reports    Researches and Applications   
   
Papers and Reports
965 Variable Precision Based Optimal Scale Combinations in Generalized Multi-scale Decision Systems
NIU Dongran, WU Weizhi, LI Tongjun
To solve the problems of knowledge representation and knowledge acquisition in generalized multi-scale decision systems, optimal scale combination selections based on dual probabilistic rough set model in generalized multi-scale decision systems are discussed. Notions of β lower approximation optimal scale combination, β upper approximation optimal scale combination, β belief distribution optimal scale combination and β plausibility distribution optimal scale combination in generalized multi-scale decision systems are defined and their properties are examined. Finally, relationships among different notions of optimal scale combinations in generalized multi-scale decision systems are analyzed. It is proved that for some special thresholds, β lower approximation optimal scale combination is equivalent to the maximum distribution optimal scale combination, whereas β upper approximation optimal scale combination is equivalent to the generalized decision optimal scale combination.
2019 Vol. 32 (11): 965-974 [Abstract] ( 473 ) [HTML 1KB] [ PDF 602KB] ( 325 )
975 Rough Entropy and Knowledge Granularity Based on Cartesian Product of Double Universes
DENG Qie, ZHANG Xianyong, YANG Jilin, CHEN Shuai
According to the Cartesian product,the covering space on two-dimensional direct product universe is constructed, and the corresponding rough entropy and knowledge granularity are investigated. Firstly, two single-universe covering spaces are induced by double-universes approximate space, and two-dimensional covering space is constructed.Then, the rough entropy and knowledge granularity based on double-universes are positioned at a single-universe covering space, and the two measures are determined in covering spaces on symmetrical single-universe and compositive two-dimensional universe by structural simulation and granular replacement. For the three sets of double metrics, the double-metrics sum, supremum and infimum, granulation monotonicity and three-way linear combination are achieved. Finally, data simulation and simulation experiment verify the effectiveness of the measure construction and the theoretical properties.
2019 Vol. 32 (11): 975-986 [Abstract] ( 324 ) [HTML 1KB] [ PDF 787KB] ( 263 )
987 Aspect Level Sentiment Analysis Based on Recurrent Neural Network with Auxiliary Memory
LIAO Xiangwen, LIN Wei, WU Yunbing, WEI Jingjing, CHEN Guolong
Aspect level sentiment analysis employs information of terms to extract features from a sentence, and it cannot utilize information of both aspects and terms simultaneously. Therefore, the model performance is low. Aiming at this problem, an aspect level sentiment analysis based on recurrent neural network with auxiliary memory is proposed. Deep bidirectional long short term memory(DBLSTM) and positional information of words are exploited to build position-weighted memory. The attention mechanism is combined with aspect terms to build aspect memory, and with position-weighted memory and aspect memory to input a multi-layer gated recurrent unit. Then, sentimental features of the aspect are obtained. Finally, sentimental polarity is identified by the normalized function. Experimental results show that the proposed method achieves better results on three public datasets with high effectiveness.
2019 Vol. 32 (11): 987-996 [Abstract] ( 379 ) [HTML 1KB] [ PDF 879KB] ( 306 )
997 Aspect Level Sentiment Classification with Multiple-Head Attention Memory Network
ZHANG Xingsheng, GAO Teng
A fine-grained sentiment classification task is to identify the opinion words with the highest degree of correlation with target words and classify the emotional polarity in the text. A deep memory network with multiple-head attention mechanism for aspect level sentiment classification is introduced. The word embedding vector of the text is stored in the memory component, and the multi-head attention mechanism is employed to simultaneously model the overall semantics of the text and the object-related semantics among the multiple feature spaces. A feedforward network layer is applied to integrate the information in multiple feature spaces as a classification feature. Experiments on SemEval-2014 dataset and the extended dataset show that the proposed method is beneficial to alleviate the selective preference of the model.
2019 Vol. 32 (11): 997-1005 [Abstract] ( 462 ) [HTML 1KB] [ PDF 711KB] ( 911 )
1006 Instance-Level Object Detection Algorithm Fusing Adversarial Learning Strategies
QIN Runnan, WANG Rui
Existing instance-level object detection algorithms based on deep learning achieve a poor detection effect on occluded objects. To solve the problem, an improved adversarial generated region-based fully convolutional networks(AGR-FCN) with the training strategy of adversarial learning is proposed. The original fully convolutional networks(R-FCN) is regarded as a fiducial frame, and adversarial mask dropout network(AMDN) is designed based on the trained R-FCN to generate occlusion features for training samples. Through the training strategy of adversarial learning between R-FCN and AMDN, the learning ability of R-FCN to the features of occluded objects is improved, and its overall instance-level object detection performance is optimized. Experiments on GMU Kitchen dataset and BHGI dataset show that AGR-FCN algorithm achieves good detection accuracy in complex and changeable unstructured environments, such as randomly varying illumination, scale, focal ratio, angle and attitude and occlusion.
2019 Vol. 32 (11): 1006-1013 [Abstract] ( 372 ) [HTML 1KB] [ PDF 962KB] ( 315 )
Researches and Applications
1014 Predicting Popularity of Online Contents via Graph Attention Based Spatial-Temporal Neural Networ
BAO Peng, XU Hao
The existing methods for predicting the popularity of online contents ignore the structural and temporal characteristics in the dynamic process of information cascades. To address this problem, a graph attention based spatial-temporal neural network(GAST-Net) is proposed to predict the popularity of online contents. The graph attention mechanism is adopted to learn the representation of cascade structure of online contents. Then, a temporal convolutional network is employed to capture the temporal features of information cascade. Finally, the popularity of online contents is mapped through a fully convolutional layer. Experimental results on datasets of Sina Weibo and American Physical Society demonstrate that GAST-Net model consistently outperforms the state-of-the-art methods.
2019 Vol. 32 (11): 1014-1021 [Abstract] ( 544 ) [HTML 1KB] [ PDF 603KB] ( 478 )
1022 Fast Video Super-Resolution Reconstruction Method Based on Motion Feature Fusion
FU Lihua, SUN Xiaowei, ZHAO Yu, LI Zonggang, HUANG Jialiang, WANG Luyuan
Video super-resolution reconstruction methods based on deep learning are often faced with the problems of long time consumption or low accuracy. A video super-resolution reconstruction method based on deep residual network is proposed. It reconstructs videos with high accuracy quickly and meets the real-time requirements for low-resolution videos. Firstly, the adaptive key frame discrimination subnet is utilized to adaptively identify key frames from the video. Then, the reconstruction results of the key frames are obtained by the high precision reconstruction subnet. For non-key frames, the reconstruction results are directly gained based on the features obtained by fusing the features of the corresponding key frame and the motion estimation features between the non-key frame and the adjacent key frame. Experiments on open datasets show that videos are fast reconstructed by the proposed method with high accuracy and robustness.
2019 Vol. 32 (11): 1022-1031 [Abstract] ( 398 ) [HTML 1KB] [ PDF 1435KB] ( 340 )
1032 Automatic Determination of Clustering Center for Clustering by Fast Search and Find of Density Peaks
WANG Wanliang, WU Fei, LÜ Chuang
Clustering center cannot be automatically selected by the algorithm of fast search and find of density peaks. To solve the problem, automatic determination of clustering centers for clustering by fast search and find of density peaks is proposed. Firstly, density and distance are normalized for the problem of uneven distribution of variables, and then the upper limit of normalized density threshold is determined by Chebyshev inequality. Standard deviation is utilized to determine the upper limit of normalized distance threshold. Finally, the upper limit of decision threshold is determined according to the decision function. Two determinants are considered comprehensively to avoid the omission of the central point selection and realize the automatic determination of the cluster centers. The experiment shows that the adaptive selection of the clustering centers of the proposed algorithm is effective with good robustness and validity.
2019 Vol. 32 (11): 1032-1041 [Abstract] ( 560 ) [HTML 1KB] [ PDF 1565KB] ( 432 )
1042 Network Representation Learning Framework Based on Adversarial Graph Convolutional Networks
CHEN Mengxue, LIU Yong
The existing network representation methods and their related variants are focused on preserving network topology structure or minimizing reconstruction error. However, data distribution of latent codes is ignored. To solve the problem, an adversarial graph convolutional networks(AGCN) is proposed. AGCN combines graph structure information and node attribute information to improve network representation learning performance, and enforces the latent codes to match a prior distribution. Moreover, an end-to-end multi-task learning framework(MTL) based on AGCN is introduced. It can perform link prediction and node classification simultaneously. The experiment shows that MTL achieves a good performance.
2019 Vol. 32 (11): 1042-1050 [Abstract] ( 366 ) [HTML 1KB] [ PDF 645KB] ( 613 )
1051 Dynamic Network Structured Pruning via Feature Coefficients of Layer Fusion
LU Haiwei, YUAN Xiaotong
Pruning is an effective way to reduce the complexity of the model. In the existing pruning methods, only the influence of the convolutional layer on the feature map is taken into account,and therefore the redundant filter cannot be determined accurately. In this paper, a dynamic network structured pruning method based on layer fusion feature coefficients is proposed. Considering the influence of convolutional layer and Batch Norm layer on the feature map, the importance of the filter is determined by multiple dynamic parameters, and the redundant filter is dynamically searched to obtain the optimal network structure. Experiments on the standard datasets of CIFAR-10 and CIFAR-100 shows that both the residual network and the lightweight network maintain high accuracy while using large pruning rates.
2019 Vol. 32 (11): 1051-1056 [Abstract] ( 385 ) [HTML 1KB] [ PDF 592KB] ( 682 )
模式识别与人工智能
 

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