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....
 
 
2021 Vol.34 Issue.1, Published 2021-01-25
Papers and Reports    Researches and Applications    Surveys and Reviews   
   
0
2021 Vol. 34 (1): 0-0 [Abstract] ( 408 ) [HTML 1KB] [ PDF 181KB] ( 544 )
Surveys and Reviews
1 Progress and Prospect of Machine Reasoning
DING Mengyuan, LAN Xuguang, PENG Ru, ZHENG Nanning

The development of machine learning algorithms are limited by the problems, such as weak generalization ability, poor robustness and lack of interpretability. In this paper, the important role of reasoning for machine learning human knowledge and logic, understanding and interpreting the world is illustrated. Firstly, the reasoning mechanism of the human brain is studied from cognitive maps, neurons and reward circuits, to brain-inspired intuitive reasoning, neural networks and reinforcement learning. Then, the current situation, progress and challenges of machine reasoning methods and their interrelationships are summarized, including intuitive reasoning, commonsense reasoning, causal reasoning and relational reasoning. Finally, the application prospects and future research directions of machine reasoning are analyzed.
2021 Vol. 34 (1): 1-13 [Abstract] ( 1310 ) [HTML 1KB] [ PDF 1218KB] ( 1285 )
Papers and Reports
14 Deep Multi-network Embedded Clustering
CHEN Rui, TANG Yongqiang, ZHANG Caixia, ZHANG Wensheng, HAO Zhifeng
Existing deep unsupervised clustering methods cannot make full use of the complementary information between the extracted features of different network structures due to the single network structure in them, and thus the clustering performance is restricted. A deep multi-network embedded clustering(DMNEC) algorithm is proposed to solve this problem. Firstly, the initialization parameters of each network are obtained by pretraining multi-network branches in an end-to-end manner. On this basis, the multi-network soft assignment is defined, then the clustering-oriented Kullback-Leibler divergence loss is established with the help of the multi-network auxiliary target distribution. The decoding network in the pretraining stage is finetuned via reconstruction loss to preserve the local structure and avoid the distortion of feature space. The weighted sum of reconstruction loss and clustering loss is optimized by stochastic gradient descent(SGD) and back propagation(BP) to jointly learn multi-network representation and cluster assignment. Experiments on four public image datasets demonstrate the superiority of the proposed algorithm.
2021 Vol. 34 (1): 14-24 [Abstract] ( 739 ) [HTML 1KB] [ PDF 1195KB] ( 804 )
25 Individual Convergence of Projected Dual Averaging Methods in Nonsmooth Convex Cases
QU Junyi, BAO Lei , TAO Qing
For convex problems, since the convergence analysis of DA(dual averaging) needs to be transformed in dual space, it is difficult to gain individual convergence. In this paper, a simple convergence analysis of DA is presented, and then it is proved that DA can attain the same optimal Ο(lnt/t) individual convergence rate as gradient descent(GD). Different from GD, the individual convergence of DA is proved to be step-size flexible by analyzing two typical step-size strategies. Furthermore, the stochastic version of the derived convergence is extended to solve large-scale machine learning problems. Experiments on L1-norm constrained hinge loss problems verify the correctness of the theoretical analysis.
2021 Vol. 34 (1): 25-32 [Abstract] ( 413 ) [HTML 1KB] [ PDF 1868KB] ( 558 )
33 Knowledge Base Completion Based on Multimodal Representation Learning
WANG Jingbin, SU Hua, LAI Xiaolian
In most learning models for knowledge graph representation, only structural knowledge between entities and relations is taken into account. Therefore, the capability of the models is limited by knowledge storage, and the completion performance of knowledge base is unstable. Existing knowledge representation methods incorporating external information mostly model for a specific kind of external modal information, leading to limited application scopes. Thus, a knowledge representation learning model, Conv-AT, is proposed. Firstly, two external modes of information, text and images, are considered, and three schemes fusing external knowledge and entities are introduced to obtain multimodal representation of entities. Secondly, the performance of convolution is enhanced and the quality of knowledge representation as well as the completion ability of the model are improved by combining the channel attention module and spatial attention module. Link prediction and triple classification experiments are conducted on public multimodal datasets, and the results show that the proposed method is superior to other methods.
2021 Vol. 34 (1): 33-43 [Abstract] ( 846 ) [HTML 1KB] [ PDF 711KB] ( 570 )
44 Robust Label Distribution Learning from a Perspective of Local Collaboration
XU Suping, SHANG Lin, ZHOU Yujie
In the most of label distribution learning(LDL)algorithms, the correlations among different labels and the overall structure of label distribution are destroyed to a certain extent. Moreover, most existing LDL algorithms mainly focus on improving the predictive performance of label distribution, while ignoring the significance of computational cost and noise robustness in practical applications. To tackle these issues, a local collaborative representation based label distribution learning algorithm (LCR-LDL) is proposed. In LCR-LDL, an unlabeled sample is treated as a collaborative representation of the local dictionary constructed by the neighborhood of the unlabeled sample, and the discriminating information of representation coefficients is utilized to reconstruct the label distribution of unlabeled sample. Experimental results on 15 real-world LDL datasets show that LCR-LDL effectively improves the predictive performance for LDL tasks with a better robustness and low computational cost.
2021 Vol. 34 (1): 44-57 [Abstract] ( 369 ) [HTML 1KB] [ PDF 891KB] ( 332 )
Researches and Applications
58 Domain Adaptation Semantic Segmentation for Urban Scene Combining Self-ensembling and Adversarial Learning
ZHANG Guimei, LU Feifei, LONG Bangyao, MIAO Jun
Aiming at the problem of high cost of urban scene label acquisition, an algorithm of domain adaptation semantic segmentation for urban scene combining self-ensembling and adversarial learning is proposed. For the inter-domain gap between source and target domains, the method of style transfer is employed to transfer the source domain into a new dataset with the style of target domain. For the problem of intra-domain gap in the target domain, the self-ensembling method is introduced and a teacher network is constructed. The teacher network is utilized to supervise and guide the student network through consistency constraints on the target domain segmentation map to reduce the intra-domain gap of the target domain and improve the segmentation accuracy. The self-training method is exploited to obtain the pseudo label of the target domain and add the pseudo label into the adversarial learning method to retrain the network and further improve the segmentation ability. Experiments on segmentation datasets verify the effectiveness of the proposed algorithm.
2021 Vol. 34 (1): 58-67 [Abstract] ( 648 ) [HTML 1KB] [ PDF 2021KB] ( 637 )
68 Adversarial Cross-Modal Retrieval Based on Association Constraint
GUO Qian, QIAN Yuhua, LIANG Xinyan
In the existing cross-modal retrieval methods, retrieval results are obtained via the subspace acquired by a certain index constraint such as distance or similarity. Since the subspaces are learned with different index constraints, retrieval results are different. To improve the robustness of common subspace, a method for adversarial cross-modal retrieval based on association constraint is proposed. The consistency of different modality features is improved by the adversarial constraint to make the discriminator in the constraint unable to distinguish which modality the subspace features come from. The association of different modality features is enhanced by the association constraint. The structural information between example pairs with the same semantics of different modalities and different semantics of the same modality is taken into account by the triple loss constraint. Experimental results on datasets show that the proposed method is more effective than other cross-modal retrieval methods.
2021 Vol. 34 (1): 68-76 [Abstract] ( 418 ) [HTML 1KB] [ PDF 690KB] ( 338 )
77 Low-Light Image Enhancement Algorithm Based on Improved Retinex-Net
OU Jiamin, HU Xiao, YANG Jiaxin
Aiming at the problems of high noise and color distortion in Retinex-Net algorithm, a low-light image enhancement algorithm based on improved Retinex-Net is proposed grounded on the decomposition-enhancement framework of Retinex-Net. Firstly, a decomposition network composed of shallow upper and lower sampling structure is designed to decompose the input image into reflection component and illumination component. In this process, the denoising loss is added to suppress the noise generated during the decomposition process. Secondly, the attention mechanism module and color loss are introduced into the enhancement network to enhance the brightness of the illumination component and meanwhile reduce the image color distortion. Finally, the reflection component and the enhanced illumination component are fused into the normal illumination image to output. The experimental results show that the proposed algorithm improves the image brightness effectively with the noise of enhanced image reduced.
2021 Vol. 34 (1): 77-86 [Abstract] ( 1597 ) [HTML 1KB] [ PDF 5339KB] ( 1060 )
87 Cross-Domain Aspect-Level Sentiment Analysis Based on Adversarial Distribution Alignment
DU Yongping, LIU Yang, HE Meng
The source domain data with rich sentiment labels is utilized to classify the aspect-level sentiment polarity for the target domain data without labels. Therefore, a cross-domain aspect-level sentiment classification model based on adversarial distribution alignment is proposed in this paper. The interactive attention of aspect words and context is employed to learn semantic relations, and the shared feature representations are learned by domain classifiers based on gradient reversal layers. The adversarial training is conducted to expand the alignment boundary of the domain distribution. And then the misclassification problem caused by fuzzy features is alleviated effectively. The experimental results on Semeval-2014 and Twitter datasets show that the performance of the proposed model is better than other classic aspect-level sentiment analysis models. The ablation experiment proves that the classification performance can be improved significantly by the strategy of capturing fuzzy features of decision boundary and expanding the distance between sample and decision boundary.
2021 Vol. 34 (1): 87-94 [Abstract] ( 578 ) [HTML 1KB] [ PDF 1791KB] ( 667 )
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