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....

2024 Vol.37 Issue.10, Published 2024-10-25 Papers and Reports    Researches and Applications    Surveys and Reviews        Surveys and Reviews 851 A Review of Multi-agent Reinforcement Learning Theory and Applications CHEN Zhuoran, LIU Zeyang, WAN Lipeng, CHEN Xingyu, ZHU Yameng, WANG Chengze, CHENG Xiang, ZHANG Ya, ZHANG Senlin, WANG Xiaohui, LAN Xuguang Reinforcement learning(RL) is a widely utilized machine learning paradigm for addressing sequential decision-making problems. Its core principle involves enabling agents to learn optimal policies iteratively through feedback derived from interactions between an agent and the environment. As the demands for computational power and data scale of practical applications continue to escalate, the transition from single-agent intelligence to collective intelligence becomes an inevitable trend in the future development of artificial intelligence. Therefore, challenges and opportunities are abundant for RL. In this paper, grounded on the concept of deep multi-agent reinforcement learning(MARL), the current theoretical dilemmas are refined and analyzed, including limited scalability, credit assignment, exploration-exploitation dilemma, non-stationarity and partial observability of information. Various solutions and their advantages and disadvantages proposed by researchers are elaborated. Typical training and learning environment of MARL and its practical applications in complex decision-making fields, such as smart city construction, gaming, robotics control and autonomous driving, are introduced. The challenges and future development direction of collaborative multi-agent reinforcement learning are summarized. 2024 Vol. 37 (10): 851-872 [Abstract] ( 115 ) [HTML 1KB] [ PDF 1775KB] ( 73 ) Papers and Reports 873 Distributed Parallel Construction Algorithm for Triadic Concepts LI Jinhai, WANG Kun, CHEN Qiangqiang As an extension of formal concept analysis, triadic concept analysis achieves significant results in both theory and applications of high-dimensional data. However, the time complexity of triadic concept generation algorithms, caused by the rapid growth of data volume, typically grows exponentially, presenting significant challenges in practical applications. Therefore, parallel algorithms are crucial. In this paper, a distributed parallel construction algorithm for triadic concepts suitable for large-scale data is proposed. First, the theories of object-attribute triadic concepts and attribute-condition triadic concepts are provided, and it is proved that all triadic concepts can be generated by merging these two types of intermediate concepts. Second, a two-stage aggregation strategy is employed to improve the resilient distributed dataset operator in the Spark framework. Consequently, the data skew problem is effectively solved and the efficiency of the proposed algorithm is significantly improved. Finally, experiments on multiple public datasets indicate that the proposed algorithm performs efficiently in generating triadic concepts for large datasets. 2024 Vol. 37 (10): 873-886 [Abstract] ( 42 ) [HTML 1KB] [ PDF 764KB] ( 52 ) 887 Underwater Image Generation Method Based on Contrastive Learning with Hard Negative Samples LIU Zijian, WANG Xingmei, CHEN Weijing, ZHANG Wansong, ZHANG Tianzi Image generation is essential to acquire scarce underwater images, and it is typically reliant on paired data. Considering the limitation of practical access to such data distributions in marine environments, a contrastive learning-based generative adversarial network(CL-GAN) is introduced to overcome the constraints of bijection in image domain. However, the model struggles to learn complex content features from noisy images due to the low quality of negative samples resulting from random sampling. To address this issue, a hard negative sample contrastive learning-based feature level GAN(HCFGAN) for underwater image generation is proposed. To improve the quality of negative samples, a hard negative sampling module(HNS) is designed to mine feature similarity between samples. The hard negative samples close to the anchor sample are incorporated into contrastive loss for complex feature learning. To ensure the complexity and comprehensiveness of negative samples, a negative sample generation module(NSG) is constructed. The adversarial training of NSG and HNS ensures the validity of hard negative samples. To enhance feature extraction capability and training stability of the model for underwater fuzzy images, a contextual feature generator and a global feature discriminator are designed. Experiments show that the underwater images generated by HCFGAN exhibit good authenticity and richness with practical value in underwater image generation. 2024 Vol. 37 (10): 887-909 [Abstract] ( 52 ) [HTML 1KB] [ PDF 11095KB] ( 46 ) Researches and Applications 910 Contrastive Learning Based Multi-view Feature Fusion Model for Aspect-Based Sentiment Analysis WU Xing, XIA Hongbin, LIU Yuan Current aspect-based sentiment analysis methods typically extract sentiment features through dependency tree and attention mechanism. These methods are susceptible to noise from irrelevant contextual information and often neglect to model the global sentiment features of sentences, making it difficult to process sentences that implicitly express sentiment. To address these problems, a contrastive learning based multi-view feature fusion model for aspect-based sentiment analysis(CLMVFF) is proposed. First, graph convolutional networks are utilized to encode information in dependency graph, constituent graph and semantic graph. The global sentiment node is constructed in each graph to learn global sentiment features while introducing external knowledge embedding to enrich sentiment features. Second, contrastive learning is exploited to mitigate the negative influence of noise. Combined with the similarity separation, the sentiment features are enhanced. Finally, the dependency graph representation, constituent graph representation, semantic graph representation and external knowledge embeddings are fused. Experimental results on three datasets demonstrate that CLMVFF achieves the improvement of performance. 2024 Vol. 37 (10): 910-922 [Abstract] ( 46 ) [HTML 1KB] [ PDF 781KB] ( 46 ) 923 Knowledge Graph Reasoning Combining Rule Inference Patterns and Fact Embedding SHAN Xiaohuan, JIANG Jiantao, CHEN Ze, SONG Baoyan Knowledge graph reasoning is an essential approach to address the incompleteness of knowledge graphs. The existing embedding-based reasoning models rely on accurate facts and suffer from poor interpretability. Rule-based reasoning models depend on the completeness of knowledge graphs, resulting in low inference performance on sparse data and an inability to express inference patterns accurately. To address these issues, a model of knowledge graph reasoning combining rule inference patterns and fact embedding(RPFE) is proposed. First, BoxE is employed as the base embedding model to achieve the embedding representation of facts. Second, the inference pattern diversity functions are designed to assist the embedding models in capturing the rules of different inference patterns, providing intuitive embedded interpretation for rule learning. Then, the fact distance consistency scoring functions are proposed to enhance the embedding representation. Finally, the rules and fact scores are optimized to compensate the lack of high-quality facts in knowledge graphs and improve the interpretability of the reasoning. Experiments on three public datasets indicate that the RPFE yields excellent performance in knowledge graph reasoning. 2024 Vol. 37 (10): 923-935 [Abstract] ( 46 ) [HTML 1KB] [ PDF 846KB] ( 54 ) 936 Few-Shot Image Classification Based on Local Contrastive Learning and Novel Class Feature Generation CHEN Ning, LIU Fan, DONG Chenwei, CHEN Zhiyu The existing image classification methods depend on large-scale manually annotated data. However, when data is limited, these methods suffer from deficiencies in both local feature representation and the number of samples. To address these issues, a method for few-shot image classification based on local contrastive learning and novel class feature generation is proposed. First, local contrastive learning is introduced to represent images as multiple local features and conduct supervised contrastive learning among these local features. Thus, the model capability to represent local features is enhanced. Second, global contrastive learning is employed to ensure the separability of the overall image features. Finally, a feature generation method is proposed to mitigate the data scarcity issue under few-shot conditions. Experiments on public datasets demonstrate the superiority of the proposed method. 2024 Vol. 37 (10): 936-946 [Abstract] ( 61 ) [HTML 1KB] [ PDF 1573KB] ( 64 )

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