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....

2022 Vol.35 Issue.5, Published 2022-05-25 Papers and Reports    Researches and Applications    Surveys and Reviews        Papers and Reports 387 Attribute Reduction in Multi-granularity Formal Decision Contexts LI Jinhai, ZHOU Xinran The existing attribute reduction methods for formal decision contexts cannot deal with multi-granularity data. Therefore, three attribute reduction methods are put forward in multi-granularity formal decision contexts to realize attribute reduction of an information system by removing the class-attribute blocks from the same category under each consistent granularity layer. Firstly, from the perspective of information granules, information entropy and conditional information entropy of the consistent granularity layer are introduced in the multi-granularity formal decision contexts to further measure the significance of attributes. Secondly, based on the average conditional information entropy and conditional information entropy in the coarsest and finest consistent formal decision contexts, the consistent granularity attribute reduction method and the coarsest and finest consistent granularity attribute reduction methods are proposed in multi-granularity formal decision contexts, and their corresponding implementation algorithms are developed. Finally, the experimental results show that the proposed attribute reduction methods are effective. In addition, it is concluded that the constraint of the consistent granularity attribute reduction method is too strict. Instead, the constraints of the coarsest and finest consistent granularity attribute reduction methods are relatively weaker. 2022 Vol. 35 (5): 387-400 [Abstract] ( 645 ) [HTML 1KB] [ PDF 647KB] ( 484 ) 401 Text Summary Generation ModelBased on Sentence Fusion and Self-Supervised Training ZOU Ao, HAO Wenning, JIN Dawei, CHEN Gang To improve the capability of sentence fusion of deep neural network text generation technique, a text summary generation model based on sentence fusion and self-supervised training is proposed. Before the model training, the training data are firstly pre-processed according to the concept of points of correspondence in the theory of sentence fusion, and thus the data can meet the needs of model training. The training of the proposed model falls into two parts. In the first stage, according to the distribution of the sentence fusion phenomenon in the dataset, the training task of the permutation language model is designed with the points of correspondence as the minimum semantic unit to enhance the ability to capture the information of the fused sentence context. In the second stage, an attention masking strategy based on the fusion information is utilized to control the information intake of the model during the text generation process to enhance the fusion ability in the text generation stage. Experiments on the open dataset show that the proposed model is superior in several evaluation metrics, including those based on statistics, deep semantics and sentence fusion ratio. 2022 Vol. 35 (5): 401-411 [Abstract] ( 525 ) [HTML 1KB] [ PDF 916KB] ( 538 ) 412 Group Recommendation Method with Nash Equilibrium Strategy and Neural Collaborative Filtering LI Lin, WANG Peipei, DU Jia, ZHOU Dong The preference fusion of group members is the central problem of group recommendation. Most of the traditional fusion strategies are single type strategy, and they cannot meet the overall preference needs of the group to some extent. Therefore, a group recommendation method with Nash equilibrium strategy and neural collaborative filtering is proposed. The nonlinear interaction of potential feature vectors between users and items is obtained through multi-layer perceptron, and then the latent factor model and multi-layer perceptron are combined to realize collaborative filtering recommendation between users and items. Furthermore, a fusion strategy based on Nash equilibrium is designed based on individual recommendation scores to ensure maximum average satisfaction of group members. Experimental results on KDD CUP dataset show that the proposed method generates better recommendation performance than the benchmark method in terms of recommendation model and fusion strategy. 2022 Vol. 35 (5): 412-421 [Abstract] ( 348 ) [HTML 1KB] [ PDF 808KB] ( 386 ) Surveys and Reviews 422 A Survey on Causal Feature Selection Based on Markov Boundary Discovery WU Xingyu, JIANG Bingbing, LÜ Shengfei, WANG Xiangyu, CHEN Qiuju, CHEN Huanhuan Causal feature selection methods,also known as Markov boundary discovery methods, select features by learning the Markov boundary(MB) of the target variable. Hence, causal feature selection methods possess better interpretability and robustness than the traditional methods. In this paper, the existing causal feature selection methods are reviewed comprehensively. The methods are divided into two types, single MB discovery algorithms and multiple MB discovery algorithms. Based on the development history of each type, the typical algorithms as well as the recent advances are introduced in detail, and the accuracy, efficiency and data dependency of the algorithms are compared. Moreover, the extended MB discovery algorithms for special applications, including semi supervised learning, multi-label learning, multi-source learning and streaming data learning, are summarized. Finally, the current hotspots and the research directions in the future of causal feature selection are analyzed. Additionally, a toolbox for causal feature selection is developed(http://home.ustc.edu.cn/~xingyuwu/MB.html), where the commonly used packages and datasets are provided. 2022 Vol. 35 (5): 422-438 [Abstract] ( 1324 ) [HTML 1KB] [ PDF 1128KB] ( 1053 ) Researches and Applications 439 Two-Universe Multi-granularity Probability Rough Sets Based on Intuitionistic Fuzzy Relations HUANG Xinhong, ZHANG Xianyong, YANG Jilin In the complex uncertainty environment, extension factors are introduced into rough sets to enhance model robustness. Based on intuitionistic fuzzy relations, multi-granularity probability rough sets are investigated in two universes in this paper. Firstly, the intuitionistic fuzzy relations and two-universe background are utilized to model multi-granularity probabilistic rough sets, and four models and their integrated algorithms are acquired, including positive optimism, positive pessimism, inverse optimism and inverse pessimism. Then, mathematical properties of model lower and upper approximations are studied from the perspectives of set operation relation, probability parameter limitation and precision size comparison. Finally, the effectiveness and property correctness of the model are verified by a medical example, and the corresponding three-way decision making is provided. Regarding multi-granularity two-universe intuitionistic-fuzzy probability rough sets, systematicness, extension and applicability of the obtained models, algorithms and properties are confirmed in depth. 2022 Vol. 35 (5): 439-450 [Abstract] ( 379 ) [HTML 1KB] [ PDF 640KB] ( 344 ) 451 Multi-agent Cooperation Algorithm Based on Individual Gap Emotion in Sparse Reward Scenarios WANG Hao, WANG Jing, FANG Baofu To address the sparse reward problem confronted by reinforcement learning in multi-agent environment, a multi-agent cooperation algorithm based on individual gap emotion is proposed grounded on the role of emotions in human learning and decision making. The approximate joint action value function is optimized end-to-end to train individual policy, and the individual action value function of each agent is taken as an evaluation of the event. A gap emotion is generated via the gap between the predicted evaluation and the actual situation. The gap emotion model is regarded as an intrinsic motivation mechanism to generate an intrinsic emotion reward for each agent as an effective supplement to the extrinsic reward. Thus, the problem of sparse extrinsic rewards is alleviated. Moreover, the intrinsic emotional reward is task-independent and consequently it possesses some generality. The effectiveness and robustness of the proposed algorithm are verified in a multi-agent pursuit scenario with different sparsity levels. 2022 Vol. 35 (5): 451-460 [Abstract] ( 502 ) [HTML 1KB] [ PDF 2129KB] ( 324 ) 461 Clustering and Retraining Based Self-Supervised Speech Representation Learning Method ZHANG Wenlin, LIU Xuepeng, NIU Tong, YANG Xukui, QU Dan The existing self-supervised speech representation learning methods based on reconstruction are trained by restoring and rebuilding speech frames. However, the phoneme category information contained in the speech frame is underutilized. Combining self-supervised learning and noisy student training, a clustering and retraining based self-supervised speech representation learning method is proposed. Firstly, based on an initial self-supervised speech representation model (the teacher model),the pseudo-label reflecting the phoneme class information is obtained via unsupervised clustering. Secondly, the pseudo-label prediction task and the original masked frame reconstruction task are combined to retrain the speech representation model(the student model). Finally, the new student model is taken as the new teacher model to optimize pseudo-labels and representation models continually by iterating the whole clustering and retraining processes. Experimental results show that the speech representation model after clustering and retraining achieves better performance in downstream phoneme recognition and speaker recognition tasks. 2022 Vol. 35 (5): 461-471 [Abstract] ( 612 ) [HTML 1KB] [ PDF 992KB] ( 362 ) 472 Sequential Recommendation Model Based on Temporal Convolution Attention Neural Network DU Yongping, NIU Jinyu, WANG Lulin, YAN Rui Sequential recommendation task aims to dynamically model user interests based on user-item interaction records for item recommendation. In sequential recommendation models, user behaviors are usually modeled as interests. The models only consider the order of user behaviors while ignoring the time interval information between users. In this paper, the time interval information of behavior sequences is taken as an important factor for prediction. A temporal convolution attention neural network model(TCAN) is proposed. In the word embedding layer, the sequential position information and time interval information are introduced, and a temporal convolutional network is designed to model the position information to obtain user's long-term preference features. In addition, the two-layer self-attention mechanism is adopted to model the association between items in the user's short-term behavior sequence, and the time interval information is fused to obtain the user's short-term interest. Finally, the global information of the training data is introduced through pre-training to improve the model recommendation performance. Experiments on three datasets show that the proposed model effectively improves recommendation performance. 2022 Vol. 35 (5): 472-480 [Abstract] ( 720 ) [HTML 1KB] [ PDF 982KB] ( 492 ) 481 2022 Vol. 35 (5): 481-482 [Abstract] ( 340 ) [HTML 1KB] [ PDF 151KB] ( 502 )

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