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.6, Published 2025-06-25 Papers and Reports    Researches and Applications    Surveys and Reviews        Surveys and Reviews 485 The Milestone Year of New AI: Review of Hotspots in Artificial Intelligence and Intelligent Technology SHEN Tianyu, ZHANG Hui, YE Peijun, WANG Fei-Yue In 2024, a milestone year for new artificial intelligence(AI), significant breakthroughs were achieved in the field of artificial intelligence. This paper reviews the hotspots and trends in artificial intelligence from various aspects, including large models and embodied intelligence, artificial intelligence generated content(AIGC), AI agents, AI for Science(AI4S) and Science for AI(S4AI), as well as AI-related policies and platforms. With the rapid advancement of large models and AI agent technologies, the application of AI continues to expand, leading to new impacts across various industries. The construction of policies and platforms is continuously improving, and new quality productive forces are showing greater development potential. For autonomous intelligence, the "new AI", the emergence of more milestone works are being expected. 2025 Vol. 38 (6): 485-504 [Abstract] ( 230 ) [HTML 1KB] [ PDF 2794KB] ( 215 ) Papers and Reports 505 Entropy Based Optimal Scale Reducts for Consistent Generalized Multi-scale Interval-Set Decision Systems HAO Tingting, WU Weizhi, TAN Anhui To solve the problem of knowledge acquisition in generalized multi-scale interval-set decision systems, an optimal scale reduction method based on interval-set decision entropy is proposed. First, the concept of scale combinations containing zero scales is defined in the generalized multi-scale interval-set decision systems. Then, similarity relations of the object sets are constructed via conditional attribute sets generated by different scale combinations to obtain the representations of corresponding information granules. Next, with the given scale combination, the concepts of interval lower approximations, interval upper approximations, interval approximation accuracy, and interval roughness of the decision classes with respect to the conditional attribute sets are defined. Furthermore, by integrating conditional entropy with interval roughness,the concept of interval-set decision entropy is introduced along with its properties. Finally, in a consistent generalized multi-scale interval-set decision system, the concepts of optimal scale combination, entropy optimal scale combination and corresponding optimal scale reducts are defined. The equivalence between the two concepts of optimal scale combinations and the equivalence between the concepts of optimal scale reduct and entropy optimal scale reduct are proved. The calculation of an optimal scale reduct is illustrated with an example. 2025 Vol. 38 (6): 505-519 [Abstract] ( 125 ) [HTML 1KB] [ PDF 701KB] ( 135 ) 520 Acquisition of Compatible Subcontexts and Concept Lattice Compression with Minimal Information Loss ZHANG Luzhen, REN Ruisi As the data scale increases, the size of concept lattice grows exponentially. Effective concept lattice compression is a critical challenge in formal concept analysis. To compress concept lattices with their structure preserved, congruence relations on the lattice are utilized. First, the ↗ and ↙ relations in the original formal context are derived from attribute concepts and object concepts, respectively. Then, the initial sets of deletion attributes(objects) are defined, and by leveraging a pruning strategy, a compatible subcontext of the original formal context is obtained through arrow-closed relations. It is proven that when singleton sets of attributes(objects) are considered as the initial sets of deletion attributes(objects), compatible subcontexts with minimal information loss from the original formal context are obtained under arrow-closed relations. Consequently, compatible subcontexts with minimized information loss are utilized to determine congruence relations on the concept lattice, thereby compressing the concept lattice. Finally, an algorithm for determining congruence relations on concept lattice via compatible subcontexts is designed. Experiments validate the feasibility and effectiveness of the proposed algorithm. 2025 Vol. 38 (6): 520-537 [Abstract] ( 109 ) [HTML 1KB] [ PDF 699KB] ( 149 ) 538 Entropy-Driven Doubly Stochastic Matrix Based Graph Matching Neural Network CAO Shuyuan, HUANG Meixiang, LU Fuliang, TU Liangping Graph matching is intended to establish node correspondence between two graph structures. Existing methods generally neglect the critical role of node matching confidence in graph matching tasks. To fully utilize the matching confidence between nodes to control the propagation of node feature information, an entropy-driven doubly stochastic matrix based graph matching neural network(EDSGM) is proposed. Node features, edge features and the similarity functions for graph matching tasks are jointly learned by deep neural networks, and the graph matching problem is simplified using graph embedding methods. The matching entropy is constructed using the doubly stochastic matrix representing the graph matching results, and the intra-graph node embedding layer and cross-graph information embedding layer are built to form an adaptive graph embedding module based on the matching entropy. This design not only achieves dimensionality reduction and problem simplification in the feature space, but also makes the propagation of node information more efficient, thereby improving matching accuracy. Furthermore, all neural network modules in EDSGM can be trained end-to-end in a supervised manner, and the same neural network can effectively handle multiple categories of graph matching tasks. Experiments demonstrate that EDSGM achieves state-of-the-art graph matching accuracy across multiple categories on Pascal VOC, SPair-71k, and Willow ObjectClass graph datasets. The results validate that matching entropy enhances the efficiency of node feature propagation and improves graph matching accuracy. 2025 Vol. 38 (6): 538-551 [Abstract] ( 153 ) [HTML 1KB] [ PDF 7275KB] ( 145 ) Researches and Applications 552 Fine-to-Coarse Grained Causality Co-Driven Approach for Explanatory Visual Question Answering SHI Yecheng, MIAO Jiali, YU Kui Explanatory visual question answering (EVQA) generates user-friendly multimodal explanations for the reasoning process while answering visual questions. Thereby, the credibility of model inference is enhanced. However, due to the lack of effective modeling of visual regions object relations, the explanations generated by existing explanatory visual question answering (EVQA) models suffer from the problem of inconsistency between visual regions and semantics. To address this issue, a fine-to-coarse grained causality co-driven (FCGC-CoD) approach for explanatory visual question answering is proposed. First, the causal relationships of visual regions features are modeled, and the influential and supportive objects are identified to enhance the multimodal representation capability of the vision-and-language pretrained model. Then, a joint variational causal inference network is designed to strengthen the coarse-grained reasoning process through fine-grained multimodal causal representations, and thus the generation of multimodal explanations and answers is achieved. Experimental results demonstrate that FCGC-CoD enhances the visual reasoning consistency of explanations while answering questions accurately. 2025 Vol. 38 (6): 552-564 [Abstract] ( 138 ) [HTML 1KB] [ PDF 3371KB] ( 141 ) 565 Image Super-Resolution Reconstruction Method Based on Heterogeneous Attention Network ZHU Yujie, ZHAO Jianwei, LIU Jieyu, ZHOU Zhenghua Existing Transformer-based image super-resolution reconstruction methods suffer from excessive computational redundancy due to their traditional multi-head design and dense self-attention mechanisms. To address this issue, an image super-resolution reconstruction method based on heterogeneous attention network is proposed in this paper. A heterogeneous multi-head self-attention block and a partial deep convolutional feedforward network block are designed. A three-branch structure is adopted in the heterogeneous multi-head self-attention block to reduce redundant computation. One dense branch for transmitting complete information is retained, a sparse branch to filter out noise is introduced, and a channel fusion branch is incorporated to supplement high-frequency information. Meanwhile, in the partial deep convolutional feedforward network block, the similarity in the activation feature maps is utilized to perform a partial deep convolutional operation to reduce the computational cost of the existing convolutional feedforward network block. Experimental results illustrate that the proposed method achieves better reconstruction performance with less computational cost. 2025 Vol. 38 (6): 565-576 [Abstract] ( 152 ) [HTML 1KB] [ PDF 3586KB] ( 144 )

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