Pub Date : 2025-07-18DOI: 10.1109/TAI.2025.3590706
Jingyang Jia;Le Wu;Shengcai Duan;Xun Chen
Gesture recognition systems based on surface electromyography (sEMG) exhibit high accuracy in laboratory settings. However, they often underperform in real-world applications due to the occurrence of unknown gestures not encountered during training. Prototype learning methods, which learn gesture prototypes and classify unknown gestures based on distances to these prototypes, effectively reject unknown gestures. However, relying solely on global feature distances may overlook subtle variations, weakening discrimination between similar features and reducing the model’s ability to identify unknown gestures resembling known ones. To address these limitations, we propose a fine-grained method that models the probability distribution of each feature point, enabling the detection of subtle differences in partial features. Specifically, we employ normalizing flows to capture detailed information at the feature-point level. This approach enhances the model’s capacity to recognize challenging unknown gestures that partially differ from known gesture patterns. In addition, we introduce synthetic unknown gestures generated by applying slight perturbations to known samples, simulating challenging unknown scenarios. We then design a novel loss function that pulls known gestures closer together while pushing synthetic unknown gestures further apart, creating a more robust rejection model. Extensive experiments on both custom and public datasets demonstrate that our method achieves an area under the curve (AUC) of 0.988 on the custom dataset and an average AUC of 0.984 and 0.782 on the two public datasets, CapgMyo-DBc and NinaproDB5, respectively. These results indicate that the proposed method provides a robust and practical solution for reliable myoelectric control in real-world applications.
{"title":"Normalizing Flow-Based Fine-Grained Modeling for Unknown Gesture Rejection in Myoelectric Pattern Recognition","authors":"Jingyang Jia;Le Wu;Shengcai Duan;Xun Chen","doi":"10.1109/TAI.2025.3590706","DOIUrl":"https://doi.org/10.1109/TAI.2025.3590706","url":null,"abstract":"Gesture recognition systems based on surface electromyography (sEMG) exhibit high accuracy in laboratory settings. However, they often underperform in real-world applications due to the occurrence of unknown gestures not encountered during training. Prototype learning methods, which learn gesture prototypes and classify unknown gestures based on distances to these prototypes, effectively reject unknown gestures. However, relying solely on global feature distances may overlook subtle variations, weakening discrimination between similar features and reducing the model’s ability to identify unknown gestures resembling known ones. To address these limitations, we propose a fine-grained method that models the probability distribution of each feature point, enabling the detection of subtle differences in partial features. Specifically, we employ normalizing flows to capture detailed information at the feature-point level. This approach enhances the model’s capacity to recognize challenging unknown gestures that partially differ from known gesture patterns. In addition, we introduce synthetic unknown gestures generated by applying slight perturbations to known samples, simulating challenging unknown scenarios. We then design a novel loss function that pulls known gestures closer together while pushing synthetic unknown gestures further apart, creating a more robust rejection model. Extensive experiments on both custom and public datasets demonstrate that our method achieves an area under the curve (AUC) of 0.988 on the custom dataset and an average AUC of 0.984 and 0.782 on the two public datasets, CapgMyo-DBc and NinaproDB5, respectively. These results indicate that the proposed method provides a robust and practical solution for reliable myoelectric control in real-world applications.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 2","pages":"1012-1024"},"PeriodicalIF":0.0,"publicationDate":"2025-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146176031","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-18DOI: 10.1109/TAI.2025.3590692
Chongchong Jin;Yuanhao Cai;Yeyao Chen;Ting Luo;Zhouyan He;Yang Song
Depth image-based rendering (DIBR) is a common method for synthesizing virtual views to achieve smooth transitions in immersive media, but its immature technology often introduces distortions, adversely affecting visual quality. Obviously, accurately assessing the quality of synthesized views is crucial for monitoring and guiding the rendering process. To this end, this article proposes a no-reference deep learning-based quality assessment method for DIBR-synthesized views, which is primarily achieved by combining a contrastive learning feature enhancement network and a high–low frequency texture interaction network, abbreviated as CONTIN. Different from the traditional methods based on handcrafted feature extraction, the proposed method employs an end-to-end deep learning approach, fully exploiting the data characteristics and feature correlations. Specifically, to address the issue of sample expansion in existing deep learning methods, a contrastive sample database is first constructed by simulating various traditional and rendering distortions based on natural images, and training is performed on this database to obtain a contrastive learning feature enhancement network, which is used to extract contrastive features. Additionally, since contrastive learning tends to focus on learning abstract semantic-level features rather than pixel-level texture details, a wavelet transform decoupling is further applied to the synthetic distortion samples to construct a high–low frequency texture interaction network for extracting texture features. Finally, the two types of features are fused and regressed to generate the final quality score. Experimental results show that the proposed method achieves superior performance across three benchmark databases (namely, IRCCyN/IVC, IETR, andMCL-3D), with PLCC reaching 0.9404, 0.8380, and 0.9666, respectively, representing improvements of 0.0179, 0.0350, and 0.0175 higher than the existing best methods.
{"title":"Contrastive Learning Feature Enhancement and High–Low Frequency Texture Interaction Networks for DIBR-Synthesized View Quality Assessment","authors":"Chongchong Jin;Yuanhao Cai;Yeyao Chen;Ting Luo;Zhouyan He;Yang Song","doi":"10.1109/TAI.2025.3590692","DOIUrl":"https://doi.org/10.1109/TAI.2025.3590692","url":null,"abstract":"Depth image-based rendering (DIBR) is a common method for synthesizing virtual views to achieve smooth transitions in immersive media, but its immature technology often introduces distortions, adversely affecting visual quality. Obviously, accurately assessing the quality of synthesized views is crucial for monitoring and guiding the rendering process. To this end, this article proposes a no-reference deep learning-based quality assessment method for DIBR-synthesized views, which is primarily achieved by combining a contrastive learning feature enhancement network and a high–low frequency texture interaction network, abbreviated as CONTIN. Different from the traditional methods based on handcrafted feature extraction, the proposed method employs an end-to-end deep learning approach, fully exploiting the data characteristics and feature correlations. Specifically, to address the issue of sample expansion in existing deep learning methods, a contrastive sample database is first constructed by simulating various traditional and rendering distortions based on natural images, and training is performed on this database to obtain a contrastive learning feature enhancement network, which is used to extract contrastive features. Additionally, since contrastive learning tends to focus on learning abstract semantic-level features rather than pixel-level texture details, a wavelet transform decoupling is further applied to the synthetic distortion samples to construct a high–low frequency texture interaction network for extracting texture features. Finally, the two types of features are fused and regressed to generate the final quality score. Experimental results show that the proposed method achieves superior performance across three benchmark databases (namely, IRCCyN/IVC, IETR, andMCL-3D), with PLCC reaching 0.9404, 0.8380, and 0.9666, respectively, representing improvements of 0.0179, 0.0350, and 0.0175 higher than the existing best methods.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 2","pages":"986-1001"},"PeriodicalIF":0.0,"publicationDate":"2025-07-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146090126","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-14DOI: 10.1109/TAI.2025.3582067
Nikhil Laxminarayana;Nimish Mishra;Prayag Tiwari;Sahil Garg;Bikash K. Behera;Ahmed Farouk
N. Laxminarayana, N. Mishra, P. Tiwari, S. Garg, B. K. Behera, and A. Farouk, “Quantum-assisted activation for supervised learning in healthcare-based intrusion detection systems,” IEEE Transactions on Artificial Intelligence, vol. 5, no. 3, pp. 977–984, Mar. 2024.
N. Laxminarayana， N. Mishra， P. Tiwari， S. Garg， B. K. Behera， A. Farouk，“基于医疗保健的入侵检测系统中监督学习的量子辅助激活”，《IEEE人工智能学报》，第5卷，第5期。3，第977-984页，2024年3月。
{"title":"Retraction Notice: Quantum-Assisted Activation for Supervised Learning in Healthcare-Based Intrusion Detection Systems","authors":"Nikhil Laxminarayana;Nimish Mishra;Prayag Tiwari;Sahil Garg;Bikash K. Behera;Ahmed Farouk","doi":"10.1109/TAI.2025.3582067","DOIUrl":"https://doi.org/10.1109/TAI.2025.3582067","url":null,"abstract":"N. Laxminarayana, N. Mishra, P. Tiwari, S. Garg, B. K. Behera, and A. Farouk, “Quantum-assisted activation for supervised learning in healthcare-based intrusion detection systems,” IEEE Transactions on Artificial Intelligence, vol. 5, no. 3, pp. 977–984, Mar. 2024.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 1","pages":"606-606"},"PeriodicalIF":0.0,"publicationDate":"2025-07-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11080238","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145898252","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-10DOI: 10.1109/TAI.2025.3586828
Randa Boukabene;Fatima Benbouzid-Si Tayeb
Community detection is a rapidly growing field, especially for multilayer networks—systems with multiple interaction types. While these networks offer great potential, analyzing them remains complex and underexplored. Recently, researchers have turned to optimization techniques to address these challenges. However, despite diverse approaches, there’s no comprehensive study consolidating these advancements. To bridge this gap, this article provides a structured review of optimization techniques for community detection in multilayer networks, classifying methods by three criteria: resolution types, optimization types, and resolution methods. This aims to clarify the field and guide future research. This effort seeks to bring clarity to the field, offering a unified perspective on existing methods, while also providing a foundation to inspire and guide future research directions.
{"title":"Optimization for Community Detection in Multilayer Networks: A Comprehensive Review and Novel Taxonomy","authors":"Randa Boukabene;Fatima Benbouzid-Si Tayeb","doi":"10.1109/TAI.2025.3586828","DOIUrl":"https://doi.org/10.1109/TAI.2025.3586828","url":null,"abstract":"Community detection is a rapidly growing field, especially for multilayer networks—systems with multiple interaction types. While these networks offer great potential, analyzing them remains complex and underexplored. Recently, researchers have turned to optimization techniques to address these challenges. However, despite diverse approaches, there’s no comprehensive study consolidating these advancements. To bridge this gap, this article provides a structured review of optimization techniques for community detection in multilayer networks, classifying methods by three criteria: resolution types, optimization types, and resolution methods. This aims to clarify the field and guide future research. This effort seeks to bring clarity to the field, offering a unified perspective on existing methods, while also providing a foundation to inspire and guide future research directions.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 2","pages":"1185-1200"},"PeriodicalIF":0.0,"publicationDate":"2025-07-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175970","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-08DOI: 10.1109/TAI.2025.3586571
Shuokang Huang;Po-Yu Chen;Peilin Zhou;Kaihan Li;Julie A. McCann
WiFi-based human sensing is gaining popularity thanks to it not requiring additional devices and not being as intrusive as cameras. Specifically, human features can be extracted from WiFi channel state information (CSI) to recognize human activities, identities, etc. However, most previous works rely on single-task learning models for recognition (e.g., to either recognize activities OR identities solely). The lack of cross-task knowledge sharing restricts these models to task-specific features and poor generalization. Recent studies have applied multitask learning (MTL) to tackle this, but their cross-task sharing modules add vast amounts of extra parameters. Such massive parameters increase model complexity and reduce time efficiency. In this article, we propose a novel zero-parameter attention sharing transformer (ZAST) to efficiently recognize both activities and identities. In ZAST, a cross-task attention on attention (CAoA) mechanism computes the relevance of attention scores for cross-task knowledge sharing, as a new paradigm for lightweight MTL. To mitigate the perturbation caused by attention sharing, we formulate a multihead similarity loss (L-MS) for stable model training. We further equip ZAST with channelwise squeeze and excitation (CSE) that efficiently learns the channel correlations of CSI. Extensive experiments on four public datasets indicate that ZAST achieves state-of-the-art recognition performance with the lowest complexity and the highest efficiency.
{"title":"Zero-Parameter Attention Sharing Transformer for Joint Human Activity and Identity Recognition","authors":"Shuokang Huang;Po-Yu Chen;Peilin Zhou;Kaihan Li;Julie A. McCann","doi":"10.1109/TAI.2025.3586571","DOIUrl":"https://doi.org/10.1109/TAI.2025.3586571","url":null,"abstract":"WiFi-based human sensing is gaining popularity thanks to it not requiring additional devices and not being as intrusive as cameras. Specifically, human features can be extracted from WiFi channel state information (CSI) to recognize human activities, identities, etc. However, most previous works rely on single-task learning models for recognition (e.g., to either recognize activities OR identities solely). The lack of cross-task knowledge sharing restricts these models to task-specific features and poor generalization. Recent studies have applied multitask learning (MTL) to tackle this, but their cross-task sharing modules add vast amounts of extra parameters. Such massive parameters increase model complexity and reduce time efficiency. In this article, we propose a novel zero-parameter attention sharing transformer (ZAST) to efficiently recognize both activities and identities. In ZAST, a cross-task attention on attention (CAoA) mechanism computes the relevance of attention scores for cross-task knowledge sharing, as a new paradigm for lightweight MTL. To mitigate the perturbation caused by attention sharing, we formulate a multihead similarity loss (L-MS) for stable model training. We further equip ZAST with channelwise squeeze and excitation (CSE) that efficiently learns the channel correlations of CSI. Extensive experiments on four public datasets indicate that ZAST achieves state-of-the-art recognition performance with the lowest complexity and the highest efficiency.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 2","pages":"960-972"},"PeriodicalIF":0.0,"publicationDate":"2025-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146176019","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-04DOI: 10.1109/TAI.2025.3586238
Rachmad Vidya Wicaksana Putra;Muhammad Shafique
Embedded AI systems are expected to incur low power/energy consumption for solving machine learning tasks, as these systems are usually power constrained (e.g., object recognition task in autonomous mobile agents with portable batteries). These requirements can be fulfilled by spiking neural networks (SNNs), since their bio-inspired spike-based operations offer high accuracy and ultra low-power/energy computation. Currently, most of SNN architectures are derived from artificial neural networks whose neurons’ architectures and operations are different from SNNs, and/or developed without considering memory budgets from the underlying processing hardware of embedded platforms. These limitations hinder SNNs from reaching their full potential in accuracy and efficiency. Toward this, we propose SpikeNAS, a novel fast memory-aware neural architecture search (NAS) framework for SNNs that quickly finds an appropriate SNN architecture with high accuracy under the given memory budgets from targeted embedded systems. To do this, our SpikeNAS employs several key steps: analyzing the impacts of network operations on the accuracy, enhancing the network architecture to improve the learning quality, developing a fast memory-aware search algorithm, and performing quantization. The experimental results show that our SpikeNAS improves the searching time and maintains high accuracy compared to state-of-the-art while meeting the given memory budgets (e.g., 29$boldsymbol{times}$, 117$boldsymbol{times}$, and 3.7$boldsymbol{times}$ faster search for CIFAR10, CIFAR100, and TinyImageNet200, respectively, using an Nvidia RTX A6000 GPU machine), thereby quickly providing the appropriate SNN architecture for the memory-constrained embedded AI systems.
{"title":"SpikeNAS: A Fast Memory-Aware Neural Architecture Search Framework for Spiking Neural Network-Based Embedded AI Systems","authors":"Rachmad Vidya Wicaksana Putra;Muhammad Shafique","doi":"10.1109/TAI.2025.3586238","DOIUrl":"https://doi.org/10.1109/TAI.2025.3586238","url":null,"abstract":"Embedded AI systems are expected to incur low power/energy consumption for solving machine learning tasks, as these systems are usually power constrained (e.g., object recognition task in autonomous mobile agents with portable batteries). These requirements can be fulfilled by spiking neural networks (SNNs), since their bio-inspired spike-based operations offer high accuracy and ultra low-power/energy computation. Currently, most of SNN architectures are derived from artificial neural networks whose neurons’ architectures and operations are different from SNNs, and/or developed without considering memory budgets from the underlying processing hardware of embedded platforms. These limitations hinder SNNs from reaching their full potential in accuracy and efficiency. Toward this, we propose <italic>SpikeNAS</i>, a novel fast memory-aware neural architecture search (NAS) framework for SNNs that quickly finds an appropriate SNN architecture with high accuracy under the given memory budgets from targeted embedded systems. To do this, our SpikeNAS employs several key steps: analyzing the impacts of network operations on the accuracy, enhancing the network architecture to improve the learning quality, developing a fast memory-aware search algorithm, and performing quantization. The experimental results show that our SpikeNAS improves the searching time and maintains high accuracy compared to state-of-the-art while meeting the given memory budgets (e.g., 29<inline-formula><tex-math>$boldsymbol{times}$</tex-math></inline-formula>, 117<inline-formula><tex-math>$boldsymbol{times}$</tex-math></inline-formula>, and 3.7<inline-formula><tex-math>$boldsymbol{times}$</tex-math></inline-formula> faster search for CIFAR10, CIFAR100, and TinyImageNet200, respectively, using an Nvidia RTX A6000 GPU machine), thereby quickly providing the appropriate SNN architecture for the memory-constrained embedded AI systems.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 2","pages":"947-959"},"PeriodicalIF":0.0,"publicationDate":"2025-07-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175972","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-04DOI: 10.1109/TAI.2025.3585868
Himani Daulat;Krishna Chauhan;Tarun Varma
Filter bank design remains a critical challenge in signal processing, particularly in achieving high-performance metrics while maintaining computational efficiency. Current methods, including various optimization algorithms, have made strides in addressing these challenges but often need to improve in balancing perfect reconstruction (PR) and magnitude response accuracy. This research addresses these gaps by introducing the Gaussian distribution crayfish optimization algorithm (GD-COA), an enhanced version of the crayfish optimization algorithm (COA), for designing a four-channel finite impulse response (FIR) filter bank. The GD-COA formulates the design problem as a meta-heuristic optimization task, integrating PR and magnitude criteria to guide the filter design. It applies to uniform (critically sampled and oversampled) and nonuniform filter banks, accommodating various sampling rates. Our results show that GD-COA achieves significant improvements in filter bank performance. Specifically, for a critically sampled uniform filter bank, it attained a PR Error of $7.2219boldsymboltimes 10^{-16}$ and a Magnitude Response Approximation Error (MRAE) of $3.8018boldsymboltimes 10^{-16}$. In an oversampled uniform filter bank, the PR Error was $1.7321boldsymboltimes 10^{-5}$ with an MRAE of $7.2444boldsymboltimes 10^{-6}$. The algorithm yielded a PR Error of $3.2831boldsymboltimes 10^{-4}$ and an MRAE of $8.5113boldsymboltimes 10^{-5}$ for a nonuniform filter bank with a consistent sampling set. When applied to a variable filter bank with an inconsistent sampling set, the PR Error was $1.1403boldsymboltimes 10^{-4}$, and the MRAE was $2.34423boldsymboltimes 10^{-5}$. These results demonstrate the GD-COA’s effectiveness in optimizing filter coefficients, ensuring minimal reconstruction errors, and satisfactory magnitude response across various design scenarios.
{"title":"Application of Gaussian Distribution Crayfish Optimization in Adaptive FIR Filter Bank: Four-Channel Uniform and Nonuniform Designs","authors":"Himani Daulat;Krishna Chauhan;Tarun Varma","doi":"10.1109/TAI.2025.3585868","DOIUrl":"https://doi.org/10.1109/TAI.2025.3585868","url":null,"abstract":"Filter bank design remains a critical challenge in signal processing, particularly in achieving high-performance metrics while maintaining computational efficiency. Current methods, including various optimization algorithms, have made strides in addressing these challenges but often need to improve in balancing perfect reconstruction (PR) and magnitude response accuracy. This research addresses these gaps by introducing the Gaussian distribution crayfish optimization algorithm (GD-COA), an enhanced version of the crayfish optimization algorithm (COA), for designing a four-channel finite impulse response (FIR) filter bank. The GD-COA formulates the design problem as a meta-heuristic optimization task, integrating PR and magnitude criteria to guide the filter design. It applies to uniform (critically sampled and oversampled) and nonuniform filter banks, accommodating various sampling rates. Our results show that GD-COA achieves significant improvements in filter bank performance. Specifically, for a critically sampled uniform filter bank, it attained a PR Error of <inline-formula><tex-math>$7.2219boldsymboltimes 10^{-16}$</tex-math></inline-formula> and a Magnitude Response Approximation Error (MRAE) of <inline-formula><tex-math>$3.8018boldsymboltimes 10^{-16}$</tex-math></inline-formula>. In an oversampled uniform filter bank, the PR Error was <inline-formula><tex-math>$1.7321boldsymboltimes 10^{-5}$</tex-math></inline-formula> with an MRAE of <inline-formula><tex-math>$7.2444boldsymboltimes 10^{-6}$</tex-math></inline-formula>. The algorithm yielded a PR Error of <inline-formula><tex-math>$3.2831boldsymboltimes 10^{-4}$</tex-math></inline-formula> and an MRAE of <inline-formula><tex-math>$8.5113boldsymboltimes 10^{-5}$</tex-math></inline-formula> for a nonuniform filter bank with a consistent sampling set. When applied to a variable filter bank with an inconsistent sampling set, the PR Error was <inline-formula><tex-math>$1.1403boldsymboltimes 10^{-4}$</tex-math></inline-formula>, and the MRAE was <inline-formula><tex-math>$2.34423boldsymboltimes 10^{-5}$</tex-math></inline-formula>. These results demonstrate the GD-COA’s effectiveness in optimizing filter coefficients, ensuring minimal reconstruction errors, and satisfactory magnitude response across various design scenarios.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 2","pages":"931-946"},"PeriodicalIF":0.0,"publicationDate":"2025-07-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175994","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-01DOI: 10.1109/TAI.2025.3585090
Erich Malan;Valentino Peluso;Andrea Calimera;Enrico Macii
This article investigates the impact of client and server learning rates on training deep neural networks in federated learning (FL). While previous research has primarily focused on optimizing the initial values of these learning rates, we demonstrate that this approach alone is insufficient for maximizing model performance and training efficiency. To address this weakness, we propose a revised two-sided learning rate optimization strategy that integrates learning rate decay schedules as tunable variables and adjusts the learning rate configurations based on the target training budget, allowing for more effective optimization. We conduct an extensive experimental evaluation to quantify the improvements offered by our approach. The results reveal that: 1) integrating decay schedules into the tuning process leads to significant performance enhancements; and 2) the optimal configuration of client-server decay schedules is strongly influenced by the training round budget. Based on these findings, we claim that performance evaluations of new FL algorithms should extend beyond the fine-tuning of the initial learning rate values, as done in the state-of-the-art approach, and include the optimization of decay schedules according to the available training budget.
{"title":"Refined Two-Sided Learning Rate Tuning for Robust Evaluation in Federated Learning","authors":"Erich Malan;Valentino Peluso;Andrea Calimera;Enrico Macii","doi":"10.1109/TAI.2025.3585090","DOIUrl":"https://doi.org/10.1109/TAI.2025.3585090","url":null,"abstract":"This article investigates the impact of client and server learning rates on training deep neural networks in federated learning (FL). While previous research has primarily focused on optimizing the initial values of these learning rates, we demonstrate that this approach alone is insufficient for maximizing model performance and training efficiency. To address this weakness, we propose a revised two-sided learning rate optimization strategy that integrates learning rate decay schedules as tunable variables and adjusts the learning rate configurations based on the target training budget, allowing for more effective optimization. We conduct an extensive experimental evaluation to quantify the improvements offered by our approach. The results reveal that: 1) integrating decay schedules into the tuning process leads to significant performance enhancements; and 2) the optimal configuration of client-server decay schedules is strongly influenced by the training round budget. Based on these findings, we claim that performance evaluations of new FL algorithms should extend beyond the fine-tuning of the initial learning rate values, as done in the state-of-the-art approach, and include the optimization of decay schedules according to the available training budget.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 2","pages":"906-917"},"PeriodicalIF":0.0,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146176029","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-01DOI: 10.1109/TAI.2025.3584905
Junfeng Zhang;Hao Ji;Tarek Raïssi;Haoyue Yang
This article investigates the practical group consensus of type-1 and type-2 T-S fuzzy positive multiagent systems (MASs). First, a positive disturbance observer and a distributed positive compensator are proposed. A group consensus protocol is designed by integrating event-triggered mechanism, which utilizes the state information of the compensator. Some feasible conditions are addressed for practical group positive consensus in the form of linear programming (LP). The key novelties are threefold: 1) a novel positive disturbance observer and compensator framework is constructed; 2) a fuzzy positive group consensus protocol is established; and 3) LP is employed for describing the corresponding conditions. Finally, two examples are provided to verify the effectiveness of the theory findings.
{"title":"Practical Group Consensus of T-S Fuzzy Positive Multiagent Systems Using Compensative Control","authors":"Junfeng Zhang;Hao Ji;Tarek Raïssi;Haoyue Yang","doi":"10.1109/TAI.2025.3584905","DOIUrl":"https://doi.org/10.1109/TAI.2025.3584905","url":null,"abstract":"This article investigates the practical group consensus of type-1 and type-2 T-S fuzzy positive multiagent systems (MASs). First, a positive disturbance observer and a distributed positive compensator are proposed. A group consensus protocol is designed by integrating event-triggered mechanism, which utilizes the state information of the compensator. Some feasible conditions are addressed for practical group positive consensus in the form of linear programming (LP). The key novelties are threefold: 1) a novel positive disturbance observer and compensator framework is constructed; 2) a fuzzy positive group consensus protocol is established; and 3) LP is employed for describing the corresponding conditions. Finally, two examples are provided to verify the effectiveness of the theory findings.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 2","pages":"892-905"},"PeriodicalIF":0.0,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146176017","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-07-01DOI: 10.1109/TAI.2025.3585095
Cong Hu;Jiangtao Song;Xiao-Jun Wu
Semisupervised learning (SSL) is gaining attention for its intrinsic ability to extract valuable information from labeled and unlabeled data with improved performance. Recently, consistency regularization methods have gained interest due to their efficient learning procedures. However, they are confined to pseudolabel or feature representation-level perturbations, negating the benefit of having both forms in a single framework. This leads to the model remaining robust to either the pseudolabel or the feature representation. To this end, we propose pseudograph regularization (PGR) for semisupervised classification, which leverages graph-based contrastive learning to unify pseudolabels and feature embeddings in a single semisupervised framework. The model imposes graph regularization on both pseudolabels and feature embeddings of unlabeled data to retain the intrinsic geometric structure. Feature embeddings into the model impose constraints on the class probability, forcing the class probability distributions of unlabeled data subject to different perturbations to be consistent. The pseudolabels regularly optimize the embedding space’s structure through graph-based contrastive learning, which allows data with similar pseudolabels to have similar feature embeddings in latent space. PGR unifies pseudolabel and feature representation of unlabeled data to improve the ability of model to resist noise interference and generalization ability. Extensive experiments on four benchmark datasets demonstrate that PGR can generate higher quality pseudolabels for unlabeled data, and is superior to the state-of-the-art (SOTA) methods.
{"title":"PGR: Pseudograph Regularization for Semisupervised Classification","authors":"Cong Hu;Jiangtao Song;Xiao-Jun Wu","doi":"10.1109/TAI.2025.3585095","DOIUrl":"https://doi.org/10.1109/TAI.2025.3585095","url":null,"abstract":"Semisupervised learning (SSL) is gaining attention for its intrinsic ability to extract valuable information from labeled and unlabeled data with improved performance. Recently, consistency regularization methods have gained interest due to their efficient learning procedures. However, they are confined to pseudolabel or feature representation-level perturbations, negating the benefit of having both forms in a single framework. This leads to the model remaining robust to either the pseudolabel or the feature representation. To this end, we propose pseudograph regularization (PGR) for semisupervised classification, which leverages graph-based contrastive learning to unify pseudolabels and feature embeddings in a single semisupervised framework. The model imposes graph regularization on both pseudolabels and feature embeddings of unlabeled data to retain the intrinsic geometric structure. Feature embeddings into the model impose constraints on the class probability, forcing the class probability distributions of unlabeled data subject to different perturbations to be consistent. The pseudolabels regularly optimize the embedding space’s structure through graph-based contrastive learning, which allows data with similar pseudolabels to have similar feature embeddings in latent space. PGR unifies pseudolabel and feature representation of unlabeled data to improve the ability of model to resist noise interference and generalization ability. Extensive experiments on four benchmark datasets demonstrate that PGR can generate higher quality pseudolabels for unlabeled data, and is superior to the state-of-the-art (SOTA) methods.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 2","pages":"918-930"},"PeriodicalIF":0.0,"publicationDate":"2025-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146176000","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","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：