RFID has great potential for applications in integrated bio-consumer electronics such as implantable medical sensors and wearable medical devices as a bridge between bio-signal acquisition and data storage. EEPROM, an important tool for storing biometric information, has been neglected in research in related fields. This paper presents a 3 Kbits EEPROM that could be utilized for passive RFID tag ICs, developed in SMIC $0.13mu $ m EEPROM 2P6M CMOS process, boasting a chip area of $338.28~mu $ m ${times } 310.78~mu $ m. The average power consumption for read and write operations was 560 nA and $31~mu $ A respectively, which is 62% and 6.5% lower than the current state-of-the-art literature. This paper proposes a novel reading circuit that reads data using only one simple inverter, greatly saving power consumption and area compared with the traditional readout method using operational amplifiers. Furthermore, this paper proposes a power-optimized charge pump that switches the frequency divider generated by the low-frequency clock when the voltage rises to a high voltage to reduce power consumption. To fulfill the low-cost advantage, this paper places the device under the MIM capacitor using the MIM (shd) and M4 layers to separate. Thus, it is more suitable for bio-consumer electronics from a commercial perspective.
{"title":"A 3 Kbits of Low-Cost, Low-Power EEPROM Integrated Into RFID Tag Integrated Circuits Available for Bio-Consumer Electronics","authors":"De-Ming Wang;Jian-Hao Cai;Jing Wu;De-Zhi Li;Jian-Guo Hu;Qing-Hua Zhong","doi":"10.1109/TCE.2025.3564645","DOIUrl":"https://doi.org/10.1109/TCE.2025.3564645","url":null,"abstract":"RFID has great potential for applications in integrated bio-consumer electronics such as implantable medical sensors and wearable medical devices as a bridge between bio-signal acquisition and data storage. EEPROM, an important tool for storing biometric information, has been neglected in research in related fields. This paper presents a 3 Kbits EEPROM that could be utilized for passive RFID tag ICs, developed in SMIC <inline-formula> <tex-math>$0.13mu $ </tex-math></inline-formula>m EEPROM 2P6M CMOS process, boasting a chip area of <inline-formula> <tex-math>$338.28~mu $ </tex-math></inline-formula>m <inline-formula> <tex-math>${times } 310.78~mu $ </tex-math></inline-formula>m. The average power consumption for read and write operations was 560 nA and <inline-formula> <tex-math>$31~mu $ </tex-math></inline-formula>A respectively, which is 62% and 6.5% lower than the current state-of-the-art literature. This paper proposes a novel reading circuit that reads data using only one simple inverter, greatly saving power consumption and area compared with the traditional readout method using operational amplifiers. Furthermore, this paper proposes a power-optimized charge pump that switches the frequency divider generated by the low-frequency clock when the voltage rises to a high voltage to reduce power consumption. To fulfill the low-cost advantage, this paper places the device under the MIM capacitor using the MIM (shd) and M4 layers to separate. Thus, it is more suitable for bio-consumer electronics from a commercial perspective.","PeriodicalId":13208,"journal":{"name":"IEEE Transactions on Consumer Electronics","volume":"71 2","pages":"5437-5445"},"PeriodicalIF":10.9,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144867972","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2025-04-28DOI: 10.1109/TCE.2025.3564849
Wazir Zada Khan;Ayesha Siddiqa;Faisal Alanazi;Muhammad Khurram Khan
The Metaverse creates a 3D virtual environment similar to the real world, enabling immersive interactions across diverse fields such as education, healthcare, and gaming. A critical aspect of these interactions is digital identity authentication, which ensures secure and trustworthy user experiences. This paper proposes a novel Non-Fungible Token (NFT) based digital identity authentication framework for the Metaverse, leveraging blockchain technology and Elliptic Curve Cryptography (ECC) to enhance security and user trust. The framework is tested using the Automated Validation of Internet Security Protocols and Applications (AVISPA) tool, demonstrating its resilience against replay and man-in-the-middle (MITM) attacks. Our contributions include: (1) a secure NFT-based authentication mechanism, (2) a formal security analysis validating the framework’s robustness, and (3) a comprehensive discussion of practical implications. The proposed framework addresses key gaps in existing methods, offering a scalable and user-friendly solution for digital identity authentication in the Metaverse.
{"title":"NFT-Based Digital Identity Authentication Framework for the Metaverse Environments","authors":"Wazir Zada Khan;Ayesha Siddiqa;Faisal Alanazi;Muhammad Khurram Khan","doi":"10.1109/TCE.2025.3564849","DOIUrl":"https://doi.org/10.1109/TCE.2025.3564849","url":null,"abstract":"The Metaverse creates a 3D virtual environment similar to the real world, enabling immersive interactions across diverse fields such as education, healthcare, and gaming. A critical aspect of these interactions is digital identity authentication, which ensures secure and trustworthy user experiences. This paper proposes a novel Non-Fungible Token (NFT) based digital identity authentication framework for the Metaverse, leveraging blockchain technology and Elliptic Curve Cryptography (ECC) to enhance security and user trust. The framework is tested using the Automated Validation of Internet Security Protocols and Applications (AVISPA) tool, demonstrating its resilience against replay and man-in-the-middle (MITM) attacks. Our contributions include: <xref>(1)</xref> a secure NFT-based authentication mechanism, <xref>(2)</xref> a formal security analysis validating the framework’s robustness, and <xref>(3)</xref> a comprehensive discussion of practical implications. The proposed framework addresses key gaps in existing methods, offering a scalable and user-friendly solution for digital identity authentication in the Metaverse.","PeriodicalId":13208,"journal":{"name":"IEEE Transactions on Consumer Electronics","volume":"71 2","pages":"5699-5707"},"PeriodicalIF":10.9,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144868217","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The convergence of consumer electronics and sports has revolutionized how physical activities are monitored and analyzed. Devices such as smartwatches and fitness trackers collect extensive time series data for applications including activity recognition and personalized training. Large Models have emerged as powerful tools for processing such complex data. However, effectively applying these models to the temporal complexity and multimodal heterogeneity of sports data remains challenging. This paper introduces the Multi-Scale Temporal Dependency and Cooperative Feature Fusion Network (MTCF-Net), a framework leveraging the capabilities of Large Models to process multimodal time series data in sports and fitness consumer electronics. MTCF-Net integrates key components: Temporal Shift Module (TSM) and Temporal Dependency Modeling (TDM) for capturing short- and long-term dependencies, Multimodal Cooperative Feature Interaction (MCFI) for dynamic cross-modal integration, and Adaptive Feature Fusion (ADF) to prioritize task-relevant features dynamically. Extensive evaluations on the UCI-HAR and PAMAP2 datasets demonstrate MTCF-Net’s state-of-the-art performance, achieving accuracy scores of 96.44% and 98.31%, respectively. Ablation studies validate its modular design, showcasing how Large Models can enhance consumer electronics for smarter and more efficient sports applications. The model’s improved accuracy and ability enable more precise performance analysis, real-time feedback, and personalized training, thereby providing tangible benefits for both athletes and fitness enthusiasts in real-world scenarios.
{"title":"MTCF-Net: Leveraging Large Models for Multimodal Time Series Analysis in Sports and Fitness Consumer Electronics","authors":"Mingxu Lu;Te Qi;Chunlei Ci;Zhe Ren;Shuo Zhang;Yanfei Lv","doi":"10.1109/TCE.2025.3564731","DOIUrl":"https://doi.org/10.1109/TCE.2025.3564731","url":null,"abstract":"The convergence of consumer electronics and sports has revolutionized how physical activities are monitored and analyzed. Devices such as smartwatches and fitness trackers collect extensive time series data for applications including activity recognition and personalized training. Large Models have emerged as powerful tools for processing such complex data. However, effectively applying these models to the temporal complexity and multimodal heterogeneity of sports data remains challenging. This paper introduces the Multi-Scale Temporal Dependency and Cooperative Feature Fusion Network (MTCF-Net), a framework leveraging the capabilities of Large Models to process multimodal time series data in sports and fitness consumer electronics. MTCF-Net integrates key components: Temporal Shift Module (TSM) and Temporal Dependency Modeling (TDM) for capturing short- and long-term dependencies, Multimodal Cooperative Feature Interaction (MCFI) for dynamic cross-modal integration, and Adaptive Feature Fusion (ADF) to prioritize task-relevant features dynamically. Extensive evaluations on the UCI-HAR and PAMAP2 datasets demonstrate MTCF-Net’s state-of-the-art performance, achieving accuracy scores of 96.44% and 98.31%, respectively. Ablation studies validate its modular design, showcasing how Large Models can enhance consumer electronics for smarter and more efficient sports applications. The model’s improved accuracy and ability enable more precise performance analysis, real-time feedback, and personalized training, thereby providing tangible benefits for both athletes and fitness enthusiasts in real-world scenarios.","PeriodicalId":13208,"journal":{"name":"IEEE Transactions on Consumer Electronics","volume":"71 2","pages":"7304-7316"},"PeriodicalIF":10.9,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144868014","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The growing implementation of autonomous cars in intelligent transportation systems requires solid traffic forecasting and incident prevention mechanisms. Yet, there are difficulties in attaining system interoperability and user acceptability. In this research, a deep learning-based framework is suggested for traffic forecasting and prevention based on the use of a forensic method on autonomous car data. A restricted boltzmann machine derives deep, weighted features which are subsequently handled by an adaptive dilated long short-term memory model optimized by using the position updated osprey optimization algorithm. Forecasted traffic data are analyzed further to formulate mitigation strategies such as optimized path planning. Experimental results demonstrate better performance compared to the baseline methods based on various metrics, highlighting the effectiveness of the framework in improving future transportation systems and autonomous vehicle forensics.
{"title":"Autonomous Vehicle Forensics: Investigating Data Streams for Traffic Prediction and Incident Mitigation","authors":"Vivek Srivastava;Sumita Mishra;Nishu Gupta;Eid Albalawi;Shakila Basheer","doi":"10.1109/TCE.2025.3564924","DOIUrl":"https://doi.org/10.1109/TCE.2025.3564924","url":null,"abstract":"The growing implementation of autonomous cars in intelligent transportation systems requires solid traffic forecasting and incident prevention mechanisms. Yet, there are difficulties in attaining system interoperability and user acceptability. In this research, a deep learning-based framework is suggested for traffic forecasting and prevention based on the use of a forensic method on autonomous car data. A restricted boltzmann machine derives deep, weighted features which are subsequently handled by an adaptive dilated long short-term memory model optimized by using the position updated osprey optimization algorithm. Forecasted traffic data are analyzed further to formulate mitigation strategies such as optimized path planning. Experimental results demonstrate better performance compared to the baseline methods based on various metrics, highlighting the effectiveness of the framework in improving future transportation systems and autonomous vehicle forensics.","PeriodicalId":13208,"journal":{"name":"IEEE Transactions on Consumer Electronics","volume":"71 1","pages":"1211-1218"},"PeriodicalIF":4.3,"publicationDate":"2025-04-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144323128","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The adoption of consumer unmanned aerial vehicles (UAVs) for logistics transportation is a key aspect of the emerging low-altitude economy, yet it faces significant challenges. Urban environments introduce complex obstacles, including dense buildings and unpredictable wind disturbances, while existing control methods struggle to balance path-following accuracy, disturbance rejection, and communication efficiency. To meet these demands, this paper proposes a quantized fuzzy learning path-following control for consumer UAVs. Firstly, a hysteresis quantized fuzzy disturbance observer (HQFDO) is proposed where the disturbances are approximately estimated by a neural network. Notably, a hysteresis quantizer is employed to reduce the communication bandwidth occupation by discretizing disturbance observations. Subsequently, a distributed velocity controller and a heading angle controller are designed to tackle the geometric and dynamic tasks separately. Specifically, the velocity controller introduces a projective arc length error to mitigate inefficiencies and safety risks associated with frequent acceleration and deceleration switches. Compared to conventional techniques, the proposed approach improves transient performance, enhances path attractivity, and optimizes communication resource utilization. Theoretical stability analysis is provided, and simulations validate the effectiveness of the proposed control strategy.
{"title":"Efficient Path-Following for Urban Logistics: A Fuzzy Control Strategy for Consumer UAVs Under Disturbance Constraints","authors":"Xingling Shao;Jun Du;Yi Xia;Zekai Zhang;Xiangwang Hou;Mérouane Debbah","doi":"10.1109/TCE.2025.3564412","DOIUrl":"https://doi.org/10.1109/TCE.2025.3564412","url":null,"abstract":"The adoption of consumer unmanned aerial vehicles (UAVs) for logistics transportation is a key aspect of the emerging low-altitude economy, yet it faces significant challenges. Urban environments introduce complex obstacles, including dense buildings and unpredictable wind disturbances, while existing control methods struggle to balance path-following accuracy, disturbance rejection, and communication efficiency. To meet these demands, this paper proposes a quantized fuzzy learning path-following control for consumer UAVs. Firstly, a hysteresis quantized fuzzy disturbance observer (HQFDO) is proposed where the disturbances are approximately estimated by a neural network. Notably, a hysteresis quantizer is employed to reduce the communication bandwidth occupation by discretizing disturbance observations. Subsequently, a distributed velocity controller and a heading angle controller are designed to tackle the geometric and dynamic tasks separately. Specifically, the velocity controller introduces a projective arc length error to mitigate inefficiencies and safety risks associated with frequent acceleration and deceleration switches. Compared to conventional techniques, the proposed approach improves transient performance, enhances path attractivity, and optimizes communication resource utilization. Theoretical stability analysis is provided, and simulations validate the effectiveness of the proposed control strategy.","PeriodicalId":13208,"journal":{"name":"IEEE Transactions on Consumer Electronics","volume":"71 2","pages":"7117-7128"},"PeriodicalIF":10.9,"publicationDate":"2025-04-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144867659","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2025-04-24DOI: 10.1109/TCE.2025.3564176
Jinxi He;Yong Luo;Yapeng Dong;Chengxiang Wang;Bingjie Chen;Bo Yang;Qingsong Liu;Zechen Li
Chinese Tuina, a manual massage therapy in traditional Chinese medicine, can be further enhanced with appropriate electrical stimulation to achieve improved physical therapy effects. However, as the several key parameters (i.e., duration, contact area and force) in Tuina manipulation treatment change, the stimulation current may become unstable and result in discomfort to the patient. We developed a portable physiotherapy instrument to assist in Tuina physiotherapy. In our design, multi-electrodes are introduced to expand depth and range of stimulation. Meanwhile, a current control method is proposed to stabilize the stimulation current during the variation of the Tuina physiotherapist’s manipulation. The Euclidean Distance about average peak-to-peak value of the ideal output and our method output voltage in a single cycle is only 1.5899, this value represents the voltage data in volts (V) after being amplified by a factor of 10 by the oscilloscope. Compared with other control algorithms, the proposed method reduces the Euclidean distance by at least 38.960%. Furthermore, the proposed method can effectively decrease the current change caused by the changes of the physiotherapist’s manipulation, thus reducing the patient’s discomfort during treatment.
{"title":"Design of Portable Physiotherapy Instrument With Stable Stimulation Current for Assisting in Tuina","authors":"Jinxi He;Yong Luo;Yapeng Dong;Chengxiang Wang;Bingjie Chen;Bo Yang;Qingsong Liu;Zechen Li","doi":"10.1109/TCE.2025.3564176","DOIUrl":"https://doi.org/10.1109/TCE.2025.3564176","url":null,"abstract":"Chinese Tuina, a manual massage therapy in traditional Chinese medicine, can be further enhanced with appropriate electrical stimulation to achieve improved physical therapy effects. However, as the several key parameters (i.e., duration, contact area and force) in Tuina manipulation treatment change, the stimulation current may become unstable and result in discomfort to the patient. We developed a portable physiotherapy instrument to assist in Tuina physiotherapy. In our design, multi-electrodes are introduced to expand depth and range of stimulation. Meanwhile, a current control method is proposed to stabilize the stimulation current during the variation of the Tuina physiotherapist’s manipulation. The Euclidean Distance about average peak-to-peak value of the ideal output and our method output voltage in a single cycle is only 1.5899, this value represents the voltage data in volts (V) after being amplified by a factor of 10 by the oscilloscope. Compared with other control algorithms, the proposed method reduces the Euclidean distance by at least 38.960%. Furthermore, the proposed method can effectively decrease the current change caused by the changes of the physiotherapist’s manipulation, thus reducing the patient’s discomfort during treatment.","PeriodicalId":13208,"journal":{"name":"IEEE Transactions on Consumer Electronics","volume":"71 2","pages":"5425-5436"},"PeriodicalIF":10.9,"publicationDate":"2025-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144867664","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2025-04-24DOI: 10.1109/TCE.2025.3563895
Ru Peng;Xingyu Chen;Xuguang Lan
Deep neural networks have achieved remarkable success in various computer vision tasks. However, in real-world applications, such as the Internet of Things (IoT), these models often struggle due to the long-tailed data distributions. For instance, in scenarios such as Holographic Counterpart Integration in IoT-based predictive maintenance for home systems or smart repair services, common operational states are prevalent in the dataset. In contrast, rare failures, such as hardware malfunctions or system breakdowns, are represented by only a few samples. This imbalance severely impacts models, making it difficult to accurately predict rare failures, leading to costly downtime or unanticipated equipment failure. Current contrastive learning-based methods are effective at optimizing feature distributions but often overlook inter-class relationships and are highly sensitive to class imbalance, which limits their generalization ability. To address these challenges, we propose the Adversarial Mixup-based supervised contrast learning (AMCL) framework, which integrates Mixup-based data augmentation with contrastive learning and incorporates an adversarial-inspired sample policy generator. AMCL generates boundary samples via a dynamically optimized Mixup strategy to enhance inter-class relationship modeling and improve predictions on ambiguous boundaries. Furthermore, we introduce a new MixCo loss function to account for the non-one-hot distribution of Mixup-generated targets, ensuring better alignment with augmented data and improving optimization efficiency. AMCL is easy to implement and achieves a performance superior to recent approaches for long-tailed recognition across various datasets such as ImageNet-LT, iNaturalist18, CIFAR-10-LT, and CIFAR-100-LT.
{"title":"Adversarial Mixup-Based Contrast Learning for Data-Driven Predictive Maintenance in Long-Tailed Recognition","authors":"Ru Peng;Xingyu Chen;Xuguang Lan","doi":"10.1109/TCE.2025.3563895","DOIUrl":"https://doi.org/10.1109/TCE.2025.3563895","url":null,"abstract":"Deep neural networks have achieved remarkable success in various computer vision tasks. However, in real-world applications, such as the Internet of Things (IoT), these models often struggle due to the long-tailed data distributions. For instance, in scenarios such as Holographic Counterpart Integration in IoT-based predictive maintenance for home systems or smart repair services, common operational states are prevalent in the dataset. In contrast, rare failures, such as hardware malfunctions or system breakdowns, are represented by only a few samples. This imbalance severely impacts models, making it difficult to accurately predict rare failures, leading to costly downtime or unanticipated equipment failure. Current contrastive learning-based methods are effective at optimizing feature distributions but often overlook inter-class relationships and are highly sensitive to class imbalance, which limits their generalization ability. To address these challenges, we propose the Adversarial Mixup-based supervised contrast learning (AMCL) framework, which integrates Mixup-based data augmentation with contrastive learning and incorporates an adversarial-inspired sample policy generator. AMCL generates boundary samples via a dynamically optimized Mixup strategy to enhance inter-class relationship modeling and improve predictions on ambiguous boundaries. Furthermore, we introduce a new MixCo loss function to account for the non-one-hot distribution of Mixup-generated targets, ensuring better alignment with augmented data and improving optimization efficiency. AMCL is easy to implement and achieves a performance superior to recent approaches for long-tailed recognition across various datasets such as ImageNet-LT, iNaturalist18, CIFAR-10-LT, and CIFAR-100-LT.","PeriodicalId":13208,"journal":{"name":"IEEE Transactions on Consumer Electronics","volume":"71 2","pages":"5249-5258"},"PeriodicalIF":10.9,"publicationDate":"2025-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144867937","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In highly competitive market of Internet service platforms, identifying and retaining potential churners through customer churn prediction techniques is crucial for maintaining platform vitality. The sequences of interaction behaviors between customers and platforms are closely related to churn prediction results. However, existing methods focus only on capturing the temporal dependencies in dynamic behavior sequences while ignoring the correlations between different behaviors. Moreover, classical methods apply only to static data, while deep learning-based methods focus on dynamic data, neither leveraging the complementary information between static and dynamic data. To address these issues, we propose a multi-modal customer churn prediction model based on Transformer with multi-scale Time-Behavior attention, TBformer, which adaptively fuses static and dynamic data. Time-Behavior module can capture multi-scale temporal dependencies and behavioral correlations in behavioral time series across time and behavior dimensions. We perform behavior-independent multi-scale dynamic feature fusion through bidirectional connection paths. Furthermore, the multi-modal fusion module based on the attention mechanism adaptively controls the fusion weights of static and dynamic features to improve performance. Extensive experiments on two publicly available datasets, KKBox and KDD, and a private dataset, HOF, demonstrate that our TBformer achieves an average AUC of 91.2% (+2.47%), outperforming the state-of-the-art customer churn prediction methods.
{"title":"TBformer: Multi-Scale Transformer With Time-Behavior Attention for Multi-Modal Customer Churn Prediction","authors":"Yushi Li;Yunfei Tao;Ming Zhu;Ziwen Chen;Zhenyu Wen;Bideng Zhu","doi":"10.1109/TCE.2025.3563905","DOIUrl":"https://doi.org/10.1109/TCE.2025.3563905","url":null,"abstract":"In highly competitive market of Internet service platforms, identifying and retaining potential churners through customer churn prediction techniques is crucial for maintaining platform vitality. The sequences of interaction behaviors between customers and platforms are closely related to churn prediction results. However, existing methods focus only on capturing the temporal dependencies in dynamic behavior sequences while ignoring the correlations between different behaviors. Moreover, classical methods apply only to static data, while deep learning-based methods focus on dynamic data, neither leveraging the complementary information between static and dynamic data. To address these issues, we propose a multi-modal customer churn prediction model based on Transformer with multi-scale Time-Behavior attention, TBformer, which adaptively fuses static and dynamic data. Time-Behavior module can capture multi-scale temporal dependencies and behavioral correlations in behavioral time series across time and behavior dimensions. We perform behavior-independent multi-scale dynamic feature fusion through bidirectional connection paths. Furthermore, the multi-modal fusion module based on the attention mechanism adaptively controls the fusion weights of static and dynamic features to improve performance. Extensive experiments on two publicly available datasets, KKBox and KDD, and a private dataset, HOF, demonstrate that our TBformer achieves an average AUC of 91.2% (+2.47%), outperforming the state-of-the-art customer churn prediction methods.","PeriodicalId":13208,"journal":{"name":"IEEE Transactions on Consumer Electronics","volume":"71 2","pages":"3192-3203"},"PeriodicalIF":10.9,"publicationDate":"2025-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144867791","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Primary-ambient extraction (PAE) is a technique to enhance the user listening experience in spatial audio reproduction. This is achieved by extracting the primary and ambient components from the sound scene. The PAE approach of ambient phase estimation with a sparsity constraint (APES) leverages the magnitude consistency of ambient components and the sparsity of the primary components to refine the PAE performance. This approach demonstrates an improved extraction accuracy when the ambient component is relatively strong. However, APES suffers from severe extraction errors when the primary amplitudes are equal in two channels of a stereo signal, which is a common sound scene in stereo signals. In this paper, the limitations of APES are analyzed, and a novel ambient phase estimation method is proposed under the joint constraints of sparsity and independence, called APESI. This method uses the independence between the primary component and the ambient component to correct the ambient phase estimation condition. Both objective and subjective experimental results demonstrate that the proposed APESI outperforms the APES and other traditional approaches in terms of extraction accuracy and ambient spatial accuracy, especially when the primary amplitudes are equal.
{"title":"Primary-Ambient Extraction Using Ambient Phase Estimate Under Joint Sparsity and Independence Constraints for Stereo Signals","authors":"Xiyu Song;Teng Tian;Shiqi Wang;Fangzhi Yao;Hongbing Qiu;Mei Wang;Hongyan Jiang","doi":"10.1109/TCE.2025.3563989","DOIUrl":"https://doi.org/10.1109/TCE.2025.3563989","url":null,"abstract":"Primary-ambient extraction (PAE) is a technique to enhance the user listening experience in spatial audio reproduction. This is achieved by extracting the primary and ambient components from the sound scene. The PAE approach of ambient phase estimation with a sparsity constraint (APES) leverages the magnitude consistency of ambient components and the sparsity of the primary components to refine the PAE performance. This approach demonstrates an improved extraction accuracy when the ambient component is relatively strong. However, APES suffers from severe extraction errors when the primary amplitudes are equal in two channels of a stereo signal, which is a common sound scene in stereo signals. In this paper, the limitations of APES are analyzed, and a novel ambient phase estimation method is proposed under the joint constraints of sparsity and independence, called APESI. This method uses the independence between the primary component and the ambient component to correct the ambient phase estimation condition. Both objective and subjective experimental results demonstrate that the proposed APESI outperforms the APES and other traditional approaches in terms of extraction accuracy and ambient spatial accuracy, especially when the primary amplitudes are equal.","PeriodicalId":13208,"journal":{"name":"IEEE Transactions on Consumer Electronics","volume":"71 2","pages":"2806-2813"},"PeriodicalIF":10.9,"publicationDate":"2025-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144868252","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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 : 2025-04-24DOI: 10.1109/TCE.2025.3563986
Saeed Iqbal;Xiaopin Zhong;Muhammad Attique Khan;Mohammad Shabaz;Zongze Wu;Dina Abdulaziz AlHammadi;Weixiang Liu;Shabbab Ali Algamdi;Yang Li
Analyzing multi-modal medical data in the setting of uncertain healthcare situations continues to be a major topic in medical image analysis and healthcare big data. Traditional machine learning algorithms are severely hampered by inaccurate data fusion, a lack of adaptability to changing patient data, and challenges managing uncertainty. These difficulties are made worse by complicated medical images and diverse data sources, which results in less accurate diagnosis and worse-than-ideal healthcare choices. To tackle these urgent problems, this paper suggests two new approaches: Continual Learning using Progressive Neural Networks (PNNs) and Tensorized Attention Mechanism for Data Fusion. The Tensorized Attention Mechanism improves multi-modal data fusion by using dynamic, task-specific attention to improve feature alignment across modalities, and the PNNs framework uses continual learning, memory augmentation, and domain adaptation to ensure robust learning under data uncertainty. We test these methods on a variety of multi-modal datasets, such as MIMIC-IV, CheXpert, MOST, OAI, and Heart Murmur, which offer a comprehensive representation of medical data from clinical reports, chest X-rays, heart murmurs, and other heterogeneous data sources. Our experimental results show notable improvements in diagnostic performance, with notable results like a CFI of 0.10, a KR score of 90.4%, and an MMC score of 0.097, indicating superior generalization and robustness across domains. Healthcare AI applications could be revolutionized by the use of specialized losses, such as Conditional Variational Autoencoder (CVAE), Adversarial Contrastive Learning (ACL), Reciprocal Regularization, and domain adaptation losses, which are essential for preventing forgetting and guaranteeing learning stability across shifting data streams.
{"title":"Transforming Healthcare Diagnostics With Tensorized Attention and Continual Learning on Multi-Modal Data","authors":"Saeed Iqbal;Xiaopin Zhong;Muhammad Attique Khan;Mohammad Shabaz;Zongze Wu;Dina Abdulaziz AlHammadi;Weixiang Liu;Shabbab Ali Algamdi;Yang Li","doi":"10.1109/TCE.2025.3563986","DOIUrl":"https://doi.org/10.1109/TCE.2025.3563986","url":null,"abstract":"Analyzing multi-modal medical data in the setting of uncertain healthcare situations continues to be a major topic in medical image analysis and healthcare big data. Traditional machine learning algorithms are severely hampered by inaccurate data fusion, a lack of adaptability to changing patient data, and challenges managing uncertainty. These difficulties are made worse by complicated medical images and diverse data sources, which results in less accurate diagnosis and worse-than-ideal healthcare choices. To tackle these urgent problems, this paper suggests two new approaches: Continual Learning using Progressive Neural Networks (PNNs) and Tensorized Attention Mechanism for Data Fusion. The Tensorized Attention Mechanism improves multi-modal data fusion by using dynamic, task-specific attention to improve feature alignment across modalities, and the PNNs framework uses continual learning, memory augmentation, and domain adaptation to ensure robust learning under data uncertainty. We test these methods on a variety of multi-modal datasets, such as MIMIC-IV, CheXpert, MOST, OAI, and Heart Murmur, which offer a comprehensive representation of medical data from clinical reports, chest X-rays, heart murmurs, and other heterogeneous data sources. Our experimental results show notable improvements in diagnostic performance, with notable results like a CFI of 0.10, a KR score of 90.4%, and an MMC score of 0.097, indicating superior generalization and robustness across domains. Healthcare AI applications could be revolutionized by the use of specialized losses, such as Conditional Variational Autoencoder (CVAE), Adversarial Contrastive Learning (ACL), Reciprocal Regularization, and domain adaptation losses, which are essential for preventing forgetting and guaranteeing learning stability across shifting data streams.","PeriodicalId":13208,"journal":{"name":"IEEE Transactions on Consumer Electronics","volume":"71 2","pages":"3391-3412"},"PeriodicalIF":10.9,"publicationDate":"2025-04-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144868164","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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：