Fuzzy clustering is an efficient tool for unsupervised data analysis, but its performance is often degraded by redundant information and outliers. To solve this issue for boosting clustering results, we propose a novel low-rank outlier-robust fuzzy clustering approach with adaptive instrinsic structure preservation (LORFC). In this method, a new low-rank feature space that contains the global components of raw data is dynamically learned for simultaneous fuzzy clustering joint feature selection, to reduce the influence of redundant information. In addition, the outliers are sufficiently extracted and removed from the low-rank features, to ensure robust clustering performance. Moreover, the local information is also embeded into this low-rank feature space by an adaptive graph, to thoroughly capture the intrinsic struture contained in data. Based on these strategies, LORFC is capable of achieving superior and reliable clustering performance, as it is not only immune to redundant information and outliers but also aware of the global joint local structure of data. LORFC is optimized by the alternative direction multiplier method (ADMM), and its temporal complexity and theoretical convergence are analyzed. In the comprehensive experiments conducted on twelve datasets, LORFC achieves better clustering results than several state-of-the-art fuzzy clustering methods in terms of clustering accuracy (CA) and normalized mutual information (NMI). Moreover, it also performs well in the test of handling extra outliers and feature selection.
{"title":"Low-Rank Outlier-Robust Fuzzy Clustering With Adaptive Intrinsic Structure Preservation","authors":"Yingxu Wang;Long Chen;Jin Zhou;Chuanbin Zhang;Zhaoyin Shi;Guang Feng","doi":"10.1109/TETCI.2025.3616047","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3616047","url":null,"abstract":"Fuzzy clustering is an efficient tool for unsupervised data analysis, but its performance is often degraded by redundant information and outliers. To solve this issue for boosting clustering results, we propose a novel low-rank outlier-robust fuzzy clustering approach with adaptive instrinsic structure preservation (LORFC). In this method, a new low-rank feature space that contains the global components of raw data is dynamically learned for simultaneous fuzzy clustering joint feature selection, to reduce the influence of redundant information. In addition, the outliers are sufficiently extracted and removed from the low-rank features, to ensure robust clustering performance. Moreover, the local information is also embeded into this low-rank feature space by an adaptive graph, to thoroughly capture the intrinsic struture contained in data. Based on these strategies, LORFC is capable of achieving superior and reliable clustering performance, as it is not only immune to redundant information and outliers but also aware of the global joint local structure of data. LORFC is optimized by the alternative direction multiplier method (ADMM), and its temporal complexity and theoretical convergence are analyzed. In the comprehensive experiments conducted on twelve datasets, LORFC achieves better clustering results than several state-of-the-art fuzzy clustering methods in terms of clustering accuracy (CA) and normalized mutual information (NMI). Moreover, it also performs well in the test of handling extra outliers and feature selection.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"10 1","pages":"1009-1024"},"PeriodicalIF":5.3,"publicationDate":"2025-10-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146026460","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-10-14DOI: 10.1109/TETCI.2025.3616054
Sougatamoy Biswas;Anup Nandy;Asim Kumar Naskar
Human action recognition (HAR) remains a challenging topic in computer vision, attracting extensive research for applications in surveillance, sports, and human-computer interaction. Existing deep learning-based HAR methods often rely on either RGB-only inputs or global attention mechanisms, which suffer from poor generalization under occlusion, background clutter, and temporal ambiguity. Moreover, conventional methods struggle to capture fine-grained spatiotemporal dependencies due to sudden changes in motion dynamics and the lack of temporal consistency constraints in sequential modeling. To address these limitations, we propose a lightweight multimodal human action recognition framework that combines complementary cues from appearance, motion, and depth modalities. These features are integrated through a mid-level feature fusion strategy to form a unified and discriminative representation of human actions. The architecture employs an Enhanced Long-Term Recurrent Convolutional Network (E-LRCN) to model both spatial and temporal dynamics efficiently. A novel Temporal Causal Self-Attention (TCSA) module is introduced to enforce directional temporal consistency. It emphasizes recent motion context, significantly improving the discrimination of action sequences. Extensive evaluations on the KTH, UCF-101, JHMDB, and HMDB51 datasets show that the proposed framework surpasses state-of-the-art methods, achieving accuracies of 98.10%, 96.28%, 83.47%, and 77.60%, respectively. These results reflect gains of up to 1.27%, 2.08%, 2.81%, and 1.04% over the best-performing benchmark models. The proposed framework improves performance while reducing computational overhead, making it suitable for real-world human action recognition on resource-constrained platforms like Jetson Nano.
{"title":"Lightweight Multimodal Feature Fusion and Spatiotemporal Learning for Human Action Recognition on Edge Devices","authors":"Sougatamoy Biswas;Anup Nandy;Asim Kumar Naskar","doi":"10.1109/TETCI.2025.3616054","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3616054","url":null,"abstract":"Human action recognition (HAR) remains a challenging topic in computer vision, attracting extensive research for applications in surveillance, sports, and human-computer interaction. Existing deep learning-based HAR methods often rely on either RGB-only inputs or global attention mechanisms, which suffer from poor generalization under occlusion, background clutter, and temporal ambiguity. Moreover, conventional methods struggle to capture fine-grained spatiotemporal dependencies due to sudden changes in motion dynamics and the lack of temporal consistency constraints in sequential modeling. To address these limitations, we propose a lightweight multimodal human action recognition framework that combines complementary cues from appearance, motion, and depth modalities. These features are integrated through a mid-level feature fusion strategy to form a unified and discriminative representation of human actions. The architecture employs an Enhanced Long-Term Recurrent Convolutional Network (E-LRCN) to model both spatial and temporal dynamics efficiently. A novel Temporal Causal Self-Attention (TCSA) module is introduced to enforce directional temporal consistency. It emphasizes recent motion context, significantly improving the discrimination of action sequences. Extensive evaluations on the KTH, UCF-101, JHMDB, and HMDB51 datasets show that the proposed framework surpasses state-of-the-art methods, achieving accuracies of 98.10%, 96.28%, 83.47%, and 77.60%, respectively. These results reflect gains of up to 1.27%, 2.08%, 2.81%, and 1.04% over the best-performing benchmark models. The proposed framework improves performance while reducing computational overhead, making it suitable for real-world human action recognition on resource-constrained platforms like Jetson Nano.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"10 1","pages":"982-995"},"PeriodicalIF":5.3,"publicationDate":"2025-10-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146026604","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-24DOI: 10.1109/TETCI.2025.3607161
{"title":"IEEE Transactions on Emerging Topics in Computational Intelligence Information for Authors","authors":"","doi":"10.1109/TETCI.2025.3607161","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3607161","url":null,"abstract":"","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 5","pages":"C3-C3"},"PeriodicalIF":5.3,"publicationDate":"2025-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11177635","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145128546","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-09-24DOI: 10.1109/TETCI.2025.3607159
{"title":"IEEE Computational Intelligence Society Information","authors":"","doi":"10.1109/TETCI.2025.3607159","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3607159","url":null,"abstract":"","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 5","pages":"C4-C4"},"PeriodicalIF":5.3,"publicationDate":"2025-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11178057","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145128545","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-08-13DOI: 10.1109/TETCI.2025.3595684
Wenhao Ma;Yu-Chen Chang;Jie Yang;Yu-Kai Wang;Chin-Teng Lin
Multi-agent systems often require agents to collaborate with or compete against other agents with diverse goals, behaviors, or strategies. Agent modeling is essential when designing adaptive policies for intelligent machine agents in multi-agent systems, as this is the means by which the controlled agent (ego agent) understands other agents' (modeled agents) behavior and extracts their meaningful policy representations. These representations can be used to enhance the ego agent's adaptive policy which is trained by reinforcement learning. However, existing agent modeling approaches typically assume the availability of local observations from modeled agents during training or a long observation trajectory for policy adaption. To remove these constrictive assumptions and improve agent modeling performance, we devised a Contrastive Learning-based Agent Modeling (CLAM) method that relies only on the local observations from the ego agent during training and execution. With these observations, CLAM is capable of generating consistent high-quality policy representations in real time right from the beginning of each episode. We evaluated the efficacy of our approach in both cooperative and competitive multi-agent environments. The experiment results demonstrate that our approach improves reinforcement learning performance by at least 28% on cooperative and competitive tasks, which exceeds the state-of-the-art.
{"title":"Contrastive Learning-Based Agent Modeling for Deep Reinforcement Learning","authors":"Wenhao Ma;Yu-Chen Chang;Jie Yang;Yu-Kai Wang;Chin-Teng Lin","doi":"10.1109/TETCI.2025.3595684","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3595684","url":null,"abstract":"Multi-agent systems often require agents to collaborate with or compete against other agents with diverse goals, behaviors, or strategies. Agent modeling is essential when designing adaptive policies for intelligent machine agents in multi-agent systems, as this is the means by which the controlled agent (ego agent) understands other agents' (modeled agents) behavior and extracts their meaningful policy representations. These representations can be used to enhance the ego agent's adaptive policy which is trained by reinforcement learning. However, existing agent modeling approaches typically assume the availability of local observations from modeled agents during training or a long observation trajectory for policy adaption. To remove these constrictive assumptions and improve agent modeling performance, we devised a <bold>C</b>ontrastive <bold>L</b>earning-based <bold>A</b>gent <bold>M</b>odeling (<bold>CLAM</b>) method that relies only on the local observations from the ego agent during training and execution. With these observations, CLAM is capable of generating consistent high-quality policy representations in real time right from the beginning of each episode. We evaluated the efficacy of our approach in both cooperative and competitive multi-agent environments. The experiment results demonstrate that our approach improves reinforcement learning performance by at least 28% on cooperative and competitive tasks, which exceeds the state-of-the-art.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 5","pages":"3719-3726"},"PeriodicalIF":5.3,"publicationDate":"2025-08-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145128550","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-23DOI: 10.1109/TETCI.2025.3586976
{"title":"IEEE Computational Intelligence Society Information","authors":"","doi":"10.1109/TETCI.2025.3586976","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3586976","url":null,"abstract":"","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 4","pages":"C4-C4"},"PeriodicalIF":5.3,"publicationDate":"2025-07-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11095292","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144687737","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-23DOI: 10.1109/TETCI.2025.3586974
{"title":"IEEE Transactions on Emerging Topics in Computational Intelligence Information for Authors","authors":"","doi":"10.1109/TETCI.2025.3586974","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3586974","url":null,"abstract":"","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 4","pages":"C3-C3"},"PeriodicalIF":5.3,"publicationDate":"2025-07-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11095290","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144687616","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-23DOI: 10.1109/TETCI.2025.3586972
{"title":"IEEE Transactions on Emerging Topics in Computational Intelligence Publication Information","authors":"","doi":"10.1109/TETCI.2025.3586972","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3586972","url":null,"abstract":"","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 4","pages":"C2-C2"},"PeriodicalIF":5.3,"publicationDate":"2025-07-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11095272","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144687720","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-20DOI: 10.1109/TETCI.2025.3576151
Yuzhe Liu;Wenjian Luo;Yingying Qiao;Kesheng Chen;Yuhui Shi
The challenge of dynamic multimodal optimization problems (DMMOPs) lies in tracking multiple global or locally acceptable optimal solutions in environments that change over time. Typically, algorithms address these problems by treating them as static multimodal optimization problems (MMOPs) over short time intervals and employing dynamic response strategies to adapt to environmental changes. This study introduces a novel approach that transforms MMOPs into multiparty multiobjective optimization problems (MPMOPs). Subsequently, we propose a multiparty multiobjective optimization framework, i.e., MPMOP-CMA, to address DMMOPs. The algorithm is structured into four stages. The first three stages occur within a static environment and include the multiparty multiobjective optimization stage, the CMA-ES search stage, and the additional search stage. The fourth stage employs dynamic response strategies to adapt when environmental changes occur. The CEC 2022 DMMOPs benchmark test suite is used to evaluate the proposed algorithm's performance. Comparative analysis with various state-of-the-art algorithms demonstrates that the proposed method exhibits competitive performance.
{"title":"Multiparty Multiobjective Optimization for Dynamic Multimodal Optimization Problems","authors":"Yuzhe Liu;Wenjian Luo;Yingying Qiao;Kesheng Chen;Yuhui Shi","doi":"10.1109/TETCI.2025.3576151","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3576151","url":null,"abstract":"The challenge of dynamic multimodal optimization problems (DMMOPs) lies in tracking multiple global or locally acceptable optimal solutions in environments that change over time. Typically, algorithms address these problems by treating them as static multimodal optimization problems (MMOPs) over short time intervals and employing dynamic response strategies to adapt to environmental changes. This study introduces a novel approach that transforms MMOPs into multiparty multiobjective optimization problems (MPMOPs). Subsequently, we propose a multiparty multiobjective optimization framework, i.e., MPMOP-CMA, to address DMMOPs. The algorithm is structured into four stages. The first three stages occur within a static environment and include the multiparty multiobjective optimization stage, the CMA-ES search stage, and the additional search stage. The fourth stage employs dynamic response strategies to adapt when environmental changes occur. The CEC 2022 DMMOPs benchmark test suite is used to evaluate the proposed algorithm's performance. Comparative analysis with various state-of-the-art algorithms demonstrates that the proposed method exhibits competitive performance.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 6","pages":"4119-4132"},"PeriodicalIF":5.3,"publicationDate":"2025-06-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145584633","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-18DOI: 10.1109/TETCI.2025.3575420
Madhuri Dubey;Jitendra Tembhurne;Richa Makhijani
Heart disease prediction using diverse, multicenter datasets poses challenges due to data heterogeneity, privacy concerns, and non-IID (Non-Independent and Identically Distributed) data. This paper introduces FedHybrid, a novel Federated Learning (FL) framework designed to address these issues by incorporating differential privacy via a Laplace mechanism, ensuring secure and optimized model aggregation and adaptive learning rate for faster convergence across IID (Independent and Identically Distributed) and non-IID data scenarios. It effectively handles eight diverse datasets with varying sample sizes, outperforming conventional FL methods like FedAvg and FedProx while preserving patient privacy. Results show that FedHybrid with differential privacy and adaptive learning rate mechanism achieves notable improvements in both convergence and accuracy. With 2 clients, FedHybrid reaches 88.24% accuracy in just 2 communication rounds, while FedAvg and FedProx require 8 and 4 rounds, respectively. For 5 clients, FedHybrid achieves 91.6% accuracy in 5 rounds, outperforming FedAvg (89.52%) and FedProx (89.8%), both of which take 10 rounds. As the number of clients increases, FedHybrid continues to excel, reaching 85.08% accuracy with 15 clients in 20 rounds, while FedAvg and FedProx take longer with lower accuracy. FedHybrid enables multicenter institutions to train models collaboratively, efficiently, and securely. The proposed approach significantly reduces communication overhead while maintaining high accuracy, making it a robust and scalable solution for federated learning in healthcare applications, particularly for heart disease prediction with clinical data.
{"title":"Enhancing Federated Learning Through Differential Privacy: Introducing FedHybrid for Multicenter Diverse Heart Disease Datasets","authors":"Madhuri Dubey;Jitendra Tembhurne;Richa Makhijani","doi":"10.1109/TETCI.2025.3575420","DOIUrl":"https://doi.org/10.1109/TETCI.2025.3575420","url":null,"abstract":"Heart disease prediction using diverse, multicenter datasets poses challenges due to data heterogeneity, privacy concerns, and non-IID (Non-Independent and Identically Distributed) data. This paper introduces FedHybrid, a novel Federated Learning (FL) framework designed to address these issues by incorporating differential privacy via a Laplace mechanism, ensuring secure and optimized model aggregation and adaptive learning rate for faster convergence across IID (Independent and Identically Distributed) and non-IID data scenarios. It effectively handles eight diverse datasets with varying sample sizes, outperforming conventional FL methods like FedAvg and FedProx while preserving patient privacy. Results show that FedHybrid with differential privacy and adaptive learning rate mechanism achieves notable improvements in both convergence and accuracy. With 2 clients, FedHybrid reaches 88.24% accuracy in just 2 communication rounds, while FedAvg and FedProx require 8 and 4 rounds, respectively. For 5 clients, FedHybrid achieves 91.6% accuracy in 5 rounds, outperforming FedAvg (89.52%) and FedProx (89.8%), both of which take 10 rounds. As the number of clients increases, FedHybrid continues to excel, reaching 85.08% accuracy with 15 clients in 20 rounds, while FedAvg and FedProx take longer with lower accuracy. FedHybrid enables multicenter institutions to train models collaboratively, efficiently, and securely. The proposed approach significantly reduces communication overhead while maintaining high accuracy, making it a robust and scalable solution for federated learning in healthcare applications, particularly for heart disease prediction with clinical data.","PeriodicalId":13135,"journal":{"name":"IEEE Transactions on Emerging Topics in Computational Intelligence","volume":"9 6","pages":"4217-4230"},"PeriodicalIF":5.3,"publicationDate":"2025-06-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145584639","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：