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....

2025 Vol.38 Issue.1, Published 2025-01-25 Papers and Reports    Researches and Applications        1 2025 Vol. 38 (1): 1-1 [Abstract] ( 93 ) [HTML 1KB] [ PDF 168KB] ( 116 ) 94 2025 Vol. 38 (1): 94-100 [Abstract] ( 59 ) [HTML 1KB] [ PDF 248KB] ( 92 ) Papers and Reports 2 Optimal Scale Reducts for Generalized Multi-scale Multiset-Valued Decision Systems LIU Mengxin, XIE Zhenhuang, WU Weizhi, ZHU Kang The knowledge representation and the knowledge acquisition of multi-scale data are crucial research directions in multi-granularity computing. While analyzing multi-scale data, a key issue is the selection of the optimal scale combination, with the aim of choosing a suitable subsystem for the final decision. To solve the problem of knowledge acquisition from multi-scale multiset-valued data, at first, similarity relations determined by the set of objects under different scale combinations are constructed based on Hellinger distance in generalized multi-scale multiset-valued decision systems, and the information granule representation is provided. Second, the concepts of optimal scale reducts and entropy optimal scale reducts are defined in consistent generalized multi-scale multiset-valued decision systems, and the equivalence between the optimal scale reducts and the entropy optimal scale reducts is proven. In inconsistent generalized multi-scale multiset-valued decision systems, the definition of generalized decision optimal scale reducts is proposed by introducing generalized decision functions. Furthermore, by employing conditional entropies and generalized decision functions, search algorithms of entropy optimal scale reducts and generalized decision optimal scale reducts are designed. Finally, a method of constructing generalized multi-scale multiset-valued decision systems is proposed, and the experiments demonstrate the validity and rationality of the proposed algorithms. 2025 Vol. 38 (1): 2-21 [Abstract] ( 143 ) [HTML 1KB] [ PDF 1181KB] ( 119 ) 22 Multi-hop Knowledge Reasoning Based on Adversarial Reinforcement Learning CHENG Lingyun, GUO Yinzhang, LIU Qingfang To address the issues of insufficient representation of complex relationships, data sparsity, and false paths in multi-hop reasoning models within existing knowledge graph question-answering systems, a multi-hop knowledge reasoning model based on adversarial reinforcement learning is proposed. First, high-order relation vectors are decomposed to parameterize and combine entity and relation features. An attention mechanism is introduced when neighboring nodes are aggregated to assign different weights, thereby enhancing the representation ability of complex relationships. Additionally, a knowledge graph embedding framework is designed to measure the credibility of <subject entity, question, answer entity> in the embedding space. Second, multi-dimensional information is integrated into the state representation of the reinforcement learning framework to enable the Agent to make reliable decisions despite data sparsity. The generator calculates the probability of candidate entities based on state information and generates answers, while the discriminator evaluates the reasonableness of the answers and the reasoning paths. The problem of false paths is alleviated by optimizing the feedback through soft rewards and path rewards, and adversarial training is utilized to alternately optimize the generator and the discriminator. Finally, the model is applied to a multi-hop question-answering system for cloud manufacturing product design knowledge to verify its effectiveness. Comparative experiments, ablation experiments and case studies verify the effectiveness of the proposed model. 2025 Vol. 38 (1): 22-35 [Abstract] ( 107 ) [HTML 1KB] [ PDF 847KB] ( 108 ) 36 Lightweight Image Super-Resolution Reconstruction Method Based on Multi-scale Spatial Adaptive Attention Network HUANG Feng, LIU Hongwei, SHEN Ying, QIU Zhaobing, CHEN Liqiong To address the challenges of high model complexity and excessive parameter counts in existing image super-resolution(SR) reconstruction methods, a lightweight image SR reconstruction method based on multi-scale spatial adaptive attention network(MSAAN) is proposed. First, a global feature modulation module(GFM) is designed to learn global texture features. Additionally, a lightweight multi-scale feature aggregation module(MFA) is introduced to adaptively aggregate high-frequency spatial features from local to global scales. Second, the multi-scale spatial adaptive attention module(MSAA) is proposed by integrating GFM and MFA. Finally, a feature interactive gated feed-forward module(FIGFF) is incorporated to enhance the local feature extraction capability while reducing the channel redundancy. Extensive experiments demonstrate that MSAAN effectively captures more comprehensive and refined features, significantly improving reconstruction quality while maintaining a lightweight structure. 2025 Vol. 38 (1): 36-50 [Abstract] ( 166 ) [HTML 1KB] [ PDF 5233KB] ( 143 ) Researches and Applications 51 Ship Maritime Trajectory Prediction Method Integrating Data Quality Enhancement and Spatio-Temporal Information Encoding Network SHI Yue, LUO He, JIANG Ruhao, WANG Guoqiang High-precision maritime vessel trajectory prediction is crucial for reducing collision risks and enhancing search and rescue efficiency. The dynamic maritime environment renders vessel trajectory data highly complex in both temporal and spatial dimensions. Existing methods exhibit insufficient attention to the quality and movement information of vessel trajectory data, making it challenging to fully capture the spatio-temporal features and correlations effectively. To address these issues, a ship maritime trajectory prediction method integrating data quality enhancement and spatio-temporal information encoding network(DQE-STIEN) is proposed. First, based on the characteristics of vessel trajectory data, a data quality enhancement algorithm is designed by combining hash mapping classification and local outlier factor-based anomaly detection using hash values to improve the quality of problematic data. Then, a spatio-temporal information encoding network with dual encoding channels is tailored for multi-attribute vessel trajectory data to extract and integrate positional information and movement features comprehensively. Finally, the spatio-temporal associations within the data are encoded and decoded to generate complete trajectory prediction results. Experimental results on AIS datasets from five different regions demonstrate the superior performance of DQE-STIEN. Moreover, DQE-STIEN exhibits certain generalizability, making itself effective for analyzing time-series data across various fields such as energy, sales, environment and finance. 2025 Vol. 38 (1): 51-67 [Abstract] ( 108 ) [HTML 1KB] [ PDF 1446KB] ( 122 ) 68 Joint Self-Expressive Subspace Clustering Method for Dynamic Data ZHANG Hantao, ZHAO Jieyu, YE Xulun Self-expressive subspace clustering methods perform well in processing high-dimensional data and are one of the key techniques in this field. However, traditional self-expressive models typically assume that the dataset is static, and it is difficult to adapt to dynamic, continuously arriving data streams. It leads to two issues: feature heterogeneity between novel data and old data and the inclusion of unknown novel classes in newly arriving samples. To address these problems, a subspace clustering framework, joint self-expressive subspace clustering method(JSSC), is proposed. JSSC is specifically designed to handle both old and novel category samples. JSSC adapts to the continuous arrival of data streams by combining a self-expressive feature learning module with a novel category sample processing module. The proposed method effectively clusters novel category samples while maintaining strong performance on existing categories. Additionally, a deep autoencoder is utilized to learn subspace basis. Thus intuitive and interpretable representations are achieved, and the known and emerging categories are simultaneously managed through pairwise objectives and regularization terms. Experimental results on benchmark datasets show that JSSC outperforms current state-of-the-art approaches in clustering tasks, particularly excelling in handling novel categories within dynamic data. 2025 Vol. 38 (1): 68-81 [Abstract] ( 93 ) [HTML 1KB] [ PDF 4440KB] ( 128 ) 82 Self-Learning TSK Fuzzy Epilepsy Assistant Detection Algorithm Incorporating Shallow and Deep Knowledge SHI Qihuan, ZHANG Xiongtao The Takagi-Sugeno-Kang (TSK) fuzzy classifier exhibits exceptional performance in handling fuzzy information for epilepsy detection. However, Due to the complexity of epileptic electroencephalogram(EEG)signals and the diverse manifestations of seizures among patients, first-order TSK fuzzy classifiers often struggle to achieve sufficient generalization from training samples.A TSK fuzzy classifier with deep and shallow self-learning knowledge integration, namely deep-shallow mix self-learning TSK(DSMT), is proposed. In DSMT, deep rules akin to human "reflection-induction" are introduced to enhance the ability of the model to mine latent information. The commonly used teacher model in knowledge distillation is replaced by the internal knowledge of the model through a static-dynamic Siamese network structure.In the static network, shallow knowledge hidden in the outputs from different batches is employed. In the dynamic network, the outputs of the static Siamese network are recorded as deep knowledge, and deep knowledge and shallow knowledge are combined. The sensitivity of the TSK fuzzy classifier to fuzzy information is leveraged to integrate both types of knowledge. DSMT enables self-learning of the first-order TSK model and improves the adaptability of the epilepsy detection system. Additionally, an optimal temperature distillation strategy is utilized to optimize knowledge transfer efficiency. Experiments on the real epilepsy datasets, CHB-MIT, TUAB, and TUEV, verify the effectiveness of DSMT. 2025 Vol. 38 (1): 82-93 [Abstract] ( 94 ) [HTML 1KB] [ PDF 2232KB] ( 105 )

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