Saturday, May. 3, 2025 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....

2018 Vol.31 Issue.12, Published 2018-12-25 Papers and Reports    Researches and Applications        Papers and Reports 1061 Task Allocation for Distributed Self-Interested Agents FU Minglan, WANG Hao, FANG Baofu, HUANG Xiaoling In the task allocation with self-interested agents, the agents cannot cooperate effectively due to their selfishness and thus their individual revenues and system performance are decreased. To make a reasonable distribution of the utilities, a self-interested agent task allocation algorithm based on the task allocation model of a distributed self-interested agent coalitional skill game is proposed. The service agents and the task agents are self-interested. They are located in different geographic locations with different scopes of vision. The utility distribution strategies are designed for task agents to make them reasonably distribute their utilities to each required skill. The task allocation results guarantee a higher system revenue even if the agents are all self-interested. The final simulation results verify the effectiveness of the proposed algorithm and examine the impacts of the scope of vision of the self-interested agents on their individual revenues and system performance. 2018 Vol. 31 (12): 1061-1073 [Abstract] ( 485 ) [HTML 1KB] [ PDF 1005KB] ( 417 ) 1074 Semi-supervised Self-training for Multiple Standpoint Analysis in Social Events LIN Junjie, WANG Lei, MAO Wenji Existing methods for standpoint analysis mainly train standpoint classification models in a supervised or unsupervised manner. It usually needs a large number of labeled data to support the training of supervised models. In contrast, the performance of unsupervised models differs greatly from that of the supervised models. To reduce the demand of labeled data in model training, and meanwhile to ensure model performance, this paper proposes a semi-supervised self-training method for multiple standpoint analysis based on social media texts related to social events. For self-training methods, selecting and adding high-quality data to the training dataset play a key role in improving the performance of classification models during the iterative training process. The proposed method first measures the classification confidence of texts based on user-level standpoint consistency. It then leverages topic information to select high-quality texts to expand the training dataset, so as to constantly improve the performance of the model. Experimental results show that the proposed method can achieve better performance in standpoint classification compared with the representative methods in the related work as well as other semi-supervised model training methods. In addition, both the user-level standpoint consistency and topic information used in the method contribute to improve the performance of standpoint classification. 2018 Vol. 31 (12): 1074-1084 [Abstract] ( 453 ) [HTML 1KB] [ PDF 845KB] ( 494 ) 1085 Multistage Pedestrian Attribute Recognition Method Based on Improved Loss Function ZHENG Shaofei, TANG Jin, LUO Bin, WANG Xiao, WANG Wenzhong There are plenty of studies on improving the performance of pedestrian attribute recognition in video surveillance scenarios by mining the positive correlations between attributes. However, the research on negative correlations is far from enough. In this paper, a deep learning based multi-stage pedestrian attribute recognition method is proposed to simultaneously explore the positive and negative correlations between attributes. In the first stage, the loss value and the accuracy of each attribute are calculated during training. In the second stage, a new network branch is designed for the attributes with larger average loss and lower average accuracy, while other attributes remain on the original branch. All attributes are predicted by these two branches jointly. In the third stage, two new network branches with same structure as the second stage are designed to optimize the parameters and improve the performance during attribute recognition. Moreover, an improved loss function increasing the distance between positive and negative samples is proposed, and it is applied in all training stages to further improve the performance. Experiments on datasets RAP and PETA validate the promising performance of the proposed method. 2018 Vol. 31 (12): 1085-1095 [Abstract] ( 881 ) [HTML 1KB] [ PDF 1023KB] ( 1102 ) 1096 Unsupervised Feature Selection Algorithm Based on Neighborhood Preserving Learning LIU Yanfang, YE Dongyi Since the sensitivity of neighborhood method for irrelevant features is high, an unsupervised feature selection algorithm based on neighborhood preserving learning(NPL) is proposed by utilizing the reconstruction coefficient of neighborhood to maintain the original data structure. Firstly, according to the similarity of each data and its neighborhood, the similarity matrix is constructed and a low dimensional space is built by introducing a mid-matrix. Secondly, an effective feature subset is selected by the Laplace multiplier method. Finally, the proposed algorithm is compared with six state-of-the-art feature selection methods on four publicly available datasets. Experimental results show the proposed method effectively identifies the representative features. 2018 Vol. 31 (12): 1096-1102 [Abstract] ( 667 ) [HTML 1KB] [ PDF 893KB] ( 1025 ) 1103 A Method Combining Knowledge Graph and Deep Learning for Drug Discovery SANG Shengtian, YANG Zhihao, LIU Xiaoxia, WANG Lei, ZHAO Di, LIN Hongfei, WANG Jian The massive growing amount of biomedical literature brings huge challenges for data mining. In this paper, a method combining knowledge graph and deep learning is proposed to discover potential therapeutic drugs for disease of interest. Firstly, a biomedical knowledge graph is constructed with the relations extracted from biomedical literature. Then, the entities and relations of the knowledge graph are converted into low dimension continuous embeddings by knowledge graph embedding method. Finally, a recurrent neural network based drug discovery model is trained by using the known drug-disease related associations. The experimental results show that the proposed method can discover drugs for diseases and provide the drug mechanism of action. 2018 Vol. 31 (12): 1103-1110 [Abstract] ( 851 ) [HTML 1KB] [ PDF 1013KB] ( 792 ) Researches and Applications 1111 Low-Rank Representation and Matrix Completion Based Face Recognition WANG Binfu, CHEN Xiaoyun, XIAO Bingsen When the face recognition method based on regression analysis is applied to the incomplete matrix, it completes the matrix firstly before using the face recognition method. Thus, the classification performance is reduced. To solve the problem, a face recognition method based on low-rank representation and low-rank matrix completion is proposed by integrating low-rank matrix completion and low-rank representation learning into a single model. The low-rank representation coefficient matrix is computed alternately and the missing entries are recovered by minimizing the representation coefficients and matrix rank. Then, the nearest neighbor classifier is used to classify the samples. Experimental results on several open face datasets show that the proposed method effectively improves the recognition performance and reduces the error of matrix completion while the entries of the training sample matrix are randomly missing. 2018 Vol. 31 (12): 1111-1119 [Abstract] ( 555 ) [HTML 1KB] [ PDF 732KB] ( 519 ) 1120 CNN with Part-of-Speech and Attention Mechanism for Targeted Sentiment Classification DU Hui, YU Xiaoming, LIU Yue, YU Zhihua, CHENG Xueqi Targets are usually discussed together. Sentiment towards the given target may be different from the sentiment polarity of the whole text. It is necessary to focus on the related context to the target in the whole semantic scenario for targeted sentiment analysis tasks. This paper presents a targeted sentiment classification method based on convolutional neural network(CNN) with Part-of-Speech(POS) and attention mechanism. POS information is introduced into the model as a supplement to text features. Attention mechanism with respect to the given target is built based on long short term memory neural network(LSTM) modeling of the input sequence. Then, the relevant parts to the target of the input text are enhanced according to the attention and the modified sequence is input to CNN sentiment classification structure to analyze the polarity towards the given target. POS information helps to capture the context with collocation relation to the target, which will help to reduce the influence of the context with similar content or short distance but no collocation relation. LSTM and CNN modeling the input text together can be beneficial to capture semantics of the whole text and those towards the given target at the same time effectively. Experiments on SemEval2014 dataset shows the effectiveness of the model compared to attention methods based on LSTM. 2018 Vol. 31 (12): 1120-1126 [Abstract] ( 475 ) [HTML 1KB] [ PDF 786KB] ( 557 ) 1127 Attention Mechanism Based Question Entity Linking REN Chaogan, YANG Yan, JIA Zhen, TANG Huijia, YU Xiuying In question entity linking, a large amount of work in data processing and feature selection is required, cumulative errors are caused easily and the linking effect is reduced. To address the issues, an attention mechanism based encoder-decoder model for entity linking(AMEDEL) is proposed. In this model, long short-term memory network is utilized to encode the questions. Then, entity mentions and disambiguation information are generated as outputs through the decoder process by attention mechanism. Finally, these outputs are linked to the entities in knowledge base. The experiments are conducted on a dataset of questions and entities about products in automotive field. The results show that the proposed model obtains satisfactory results by only employing rare contextual information. 2018 Vol. 31 (12): 1127-1133 [Abstract] ( 571 ) [HTML 1KB] [ PDF 976KB] ( 770 ) 1134 Robust Pedestrian Detection Based on Parallel Channel Cascade Network HE Jiaojiao, ZHANG Yongping, YAO Tuozhong, LIU Ken, XIAO Jiangjian In the wide-angle field with perspective distortion, the resolution of distant pedestrian is low and there is distortion in a broad range of scales. Aiming at these problems, a robust pedestrian detection algorithm based on parallel channel cascade network is proposed. Firstly, differential information is introduced as weak supervisory information. Secondly, a new feature extraction network, channel cascade network(CCN), is proposed. On this basis, a parallel CCN is designed, and the difference map and the original map are utilized as its input. More abundant image features are fused. Finally, in the region proposal network, the distribution of pedestrians in the picture is characterized by clustering, and anchors meeting the pedestrian's characteristics are clustered. Experimental results show that the proposed algorithm is better than the standard Faster-RCNN algorithm and FPN algorithm for small-scale pedestrian detection in the presence of wide-angle field of view distortion. 2018 Vol. 31 (12): 1134-1142 [Abstract] ( 475 ) [HTML 1KB] [ PDF 1898KB] ( 531 ) 1143 Iterative Bootstrapping Attribute Knowledge Base Extension Algorithm Based on Word Co-occurrence Graph LI Zhixu, SHEN Yongxin, CHEN Jia, LIU An, ZHAO Pengpeng, ZHAO Lei Existing open information extraction methods in the attribute knowledge base extension heavily rely on deep syntax analysis or effective dictionary rules, thus the poor results in short text processing and low recall rates are produced. Therefore, an iterative bootstrapping attribute knowledge base extension algorithm based on word co-occurrence graph is proposed. The co-occurrence relationship between attribute and attribute values is employed to extend the knowledge base and a graph-based community discovery algorithm is designed to find out core nodes of the community. Finally, a model based on convolutional neural network is constructed to denoise the extraction results. Experiments on two real datasets show that the proposed method outperforms the existing ones. 2018 Vol. 31 (12): 1143-1150 [Abstract] ( 417 ) [HTML 1KB] [ PDF 826KB] ( 525 )

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