Pub Date : 2025-12-05DOI: 10.1109/TETC.2025.3633547
{"title":"IEEE Transactions on Emerging Topics in Computing Publication Information","authors":"","doi":"10.1109/TETC.2025.3633547","DOIUrl":"https://doi.org/10.1109/TETC.2025.3633547","url":null,"abstract":"","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"13 4","pages":"C2-C2"},"PeriodicalIF":5.4,"publicationDate":"2025-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11279973","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145674761","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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}
Multi-view partial multi-label learning deals with scenarios where samples contain heterogeneous features and are associated with both relevant and corrupted labels. Existing methods struggle to effectively capture label-related features through adequate feature interaction while simultaneously integrating inter- and intra-view features. To address these challenges, we propose a robust and scalable framework, Class Activation Specific Features Collaborative Network, designed to handle feature heterogeneity and facilitate comprehensive feature fusion in multi-view partial multi-label learning. The framework integrates label-specific feature extraction with collaborative information propagation through two key components: 1) View-Specific Class Activation Map, which transforms multi-view features into compact class label representations and 2) Class Information Propagation Correction, which refines and propagates accurate class label information by leveraging graph convolutional networks and transformers. Additionally, we introduce a multi-faceted loss function that promotes robust feature learning and architectural stability via consistency-based structural loss, while improving generalization through knowledge distillation. Extensive experiments on benchmark datasets demonstrate that the proposed model significantly outperforms state-of-the-art methods in multi-view partial multi-label learning tasks.
多视图部分多标签学习处理样本包含异构特征并与相关和损坏标签相关联的场景。现有的方法很难通过充分的特征交互来有效地捕获与标签相关的特征，同时集成视图间和视图内的特征。为了解决这些挑战，我们提出了一个强大且可扩展的框架，类激活特定特征协作网络，旨在处理特征异质性并促进多视图部分多标签学习中的全面特征融合。该框架通过两个关键组件将特定于标签的特征提取与协同信息传播集成在一起：1)特定于视图的类激活图（View-Specific Class Activation Map），它将多视图特征转换为紧凑的类标签表示；2)类信息传播校正（Class information propagation Correction），它利用图卷积网络和变压器精炼和传播准确的类标签信息。此外，我们引入了一个多面损失函数，通过基于一致性的结构损失促进鲁棒特征学习和架构稳定性，同时通过知识蒸馏提高泛化。在基准数据集上的大量实验表明，所提出的模型在多视图部分多标签学习任务中显著优于最先进的方法。
{"title":"Multi-View Partial Multi-Label Learning via Class Activation Specific Features Collaborative Learning","authors":"Anhui Tan;Jianhang Xu;Weiping Ding;Jiye Liang;Witold Pedrycz","doi":"10.1109/TETC.2025.3629677","DOIUrl":"https://doi.org/10.1109/TETC.2025.3629677","url":null,"abstract":"Multi-view partial multi-label learning deals with scenarios where samples contain heterogeneous features and are associated with both relevant and corrupted labels. Existing methods struggle to effectively capture label-related features through adequate feature interaction while simultaneously integrating inter- and intra-view features. To address these challenges, we propose a robust and scalable framework, Class Activation Specific Features Collaborative Network, designed to handle feature heterogeneity and facilitate comprehensive feature fusion in multi-view partial multi-label learning. The framework integrates label-specific feature extraction with collaborative information propagation through two key components: 1) View-Specific Class Activation Map, which transforms multi-view features into compact class label representations and 2) Class Information Propagation Correction, which refines and propagates accurate class label information by leveraging graph convolutional networks and transformers. Additionally, we introduce a multi-faceted loss function that promotes robust feature learning and architectural stability via consistency-based structural loss, while improving generalization through knowledge distillation. Extensive experiments on benchmark datasets demonstrate that the proposed model significantly outperforms state-of-the-art methods in multi-view partial multi-label learning tasks.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"13 4","pages":"1522-1535"},"PeriodicalIF":5.4,"publicationDate":"2025-11-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145729447","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 : 2025-11-11DOI: 10.1109/TETC.2025.3629528
Koffka Khan
Federated learning (FL) often suffers from client heterogeneity – differences in data distributions and learning behavior across clients that can degrade the global model’s performance. This paper addresses this challenge with HIFLA (Hilbert-Inspired Federated Learning via Action Principles), a novel approach that leverages variational mechanics. HIFLA formulates the federated training process as the minimization of a global action functional, yielding entropy- regularized Euler–Lagrange dynamics for client and server updates. A key innovation is the introduction of an interaction potential among client models, which mitigates divergence caused by non-i.i.d. data by coupling their updates in the action formulation. Empirically, HIFLA improves model accuracy on heterogeneous FL benchmarks, outperforming standard methods in the presence of statistical heterogeneity. It also demonstrates enhanced robustness against adversarial clients: even when a fraction of participants behave maliciously or send corrupted updates, the HIFLA-based model converges reliably with minimal performance loss. Overall, our results indicate that an action-principle-driven paradigm can effectively tackle client heterogeneity and adversarial robustness in federated learning, paving the way for more resilient and generalizable FL systems.
联邦学习（FL）经常受到客户机异构性的困扰——客户机之间数据分布和学习行为的差异会降低全局模型的性能。本文通过HIFLA （Hilbert-Inspired Federated Learning via Action Principles）解决了这一挑战，HIFLA是一种利用变分机制的新方法。HIFLA将联邦训练过程表述为全局动作函数的最小化，为客户端和服务器更新产生熵-正则化欧拉-拉格朗日动态。一个关键的创新是引入了客户模型之间的交互潜力，这减轻了由非i.d引起的分歧。数据通过在动作公式中耦合它们的更新。从经验上看，HIFLA提高了异构FL基准上的模型准确性，在存在统计异质性的情况下优于标准方法。它还展示了针对对抗性客户端的增强鲁棒性：即使一小部分参与者行为恶意或发送损坏的更新，基于hifl的模型也能以最小的性能损失可靠地收敛。总体而言，我们的研究结果表明，行动原则驱动的范式可以有效地解决联邦学习中的客户异质性和对抗性鲁棒性，为更具弹性和可泛化的FL系统铺平道路。
{"title":"HIFLA: Hilbert-Inspired Federated Learning via Action Principles","authors":"Koffka Khan","doi":"10.1109/TETC.2025.3629528","DOIUrl":"https://doi.org/10.1109/TETC.2025.3629528","url":null,"abstract":"Federated learning (FL) often suffers from client heterogeneity – differences in data distributions and learning behavior across clients that can degrade the global model’s performance. This paper addresses this challenge with HIFLA (Hilbert-Inspired Federated Learning via Action Principles), a novel approach that leverages variational mechanics. HIFLA formulates the federated training process as the minimization of a global action functional, yielding entropy- regularized Euler–Lagrange dynamics for client and server updates. A key innovation is the introduction of an <italic>interaction potential</i> among client models, which mitigates divergence caused by non-i.i.d. data by coupling their updates in the action formulation. Empirically, HIFLA improves model accuracy on heterogeneous FL benchmarks, outperforming standard methods in the presence of statistical heterogeneity. It also demonstrates enhanced robustness against adversarial clients: even when a fraction of participants behave maliciously or send corrupted updates, the HIFLA-based model converges reliably with minimal performance loss. Overall, our results indicate that an action-principle-driven paradigm can effectively tackle client heterogeneity and adversarial robustness in federated learning, paving the way for more resilient and generalizable FL systems.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"13 4","pages":"1536-1552"},"PeriodicalIF":5.4,"publicationDate":"2025-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145729326","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}
Zeroing neural network (ZNN) with variable convergence parameter has been a research hotspot recently, and it can be divided into two categories: the varying-parameter ZNN (VP-ZNN) model, and the fuzzy-parameter ZNN (FP-ZNN) model. Both of the two models have their own advantages and disadvantages. VP-ZNN models are usually efficient but not intelligent, while FP-ZNN models are usually intelligent but not efficient. Inspired by the classic proportional–integral–derivative (PID) control technology, we proposed a novel proportional-integral-parameter ZNN (PIP-ZNN) model, which is both efficient and intelligent, to solve the quaternion-valued time-varying linear matrix inequalities (QVTV-LMI) problem. Integrated with adaptive convergence parameters (ACP), the PIP-ZNN model dynamically adjusts its convergence in response to error changes and then achieves optimized performance. Compared with FP-ZNN models which are based on fuzzy logic systems, the PID-based PIP-ZNN models are simpler and more efficient. With the incorporation of a robust activation function (RAF), the PIP-ZNN model demonstrates fixed-time stability and robustness in the presence of both attenuated and constant disturbances. Theoretical analyses in this paper establish the fixed-time stability and robustness of the PIP-ZNN model, including an estimation of the upper bound of the settling-time function. Numerical experiments here validate these advanced features further, emphasizing the efficacy and excellent performance of the proposed PIP-ZNN model.
{"title":"A Novel Proportional-Integral-Parameter Zeroing Neural Network and Its Application to the Quaternion-Valued Time-Varying Linear Matrix Inequality","authors":"Jiajie Luo;Jiguang Li;Lin Xiao;Jichun Li;Wenxing Ji;William Holderbaum;Peng Qi","doi":"10.1109/TETC.2025.3629357","DOIUrl":"https://doi.org/10.1109/TETC.2025.3629357","url":null,"abstract":"Zeroing neural network (ZNN) with variable convergence parameter has been a research hotspot recently, and it can be divided into two categories: the varying-parameter ZNN (VP-ZNN) model, and the fuzzy-parameter ZNN (FP-ZNN) model. Both of the two models have their own advantages and disadvantages. VP-ZNN models are usually efficient but not intelligent, while FP-ZNN models are usually intelligent but not efficient. Inspired by the classic proportional–integral–derivative (PID) control technology, we proposed a novel proportional-integral-parameter ZNN (PIP-ZNN) model, which is both efficient and intelligent, to solve the quaternion-valued time-varying linear matrix inequalities (QVTV-LMI) problem. Integrated with adaptive convergence parameters (ACP), the PIP-ZNN model dynamically adjusts its convergence in response to error changes and then achieves optimized performance. Compared with FP-ZNN models which are based on fuzzy logic systems, the PID-based PIP-ZNN models are simpler and more efficient. With the incorporation of a robust activation function (RAF), the PIP-ZNN model demonstrates fixed-time stability and robustness in the presence of both attenuated and constant disturbances. Theoretical analyses in this paper establish the fixed-time stability and robustness of the PIP-ZNN model, including an estimation of the upper bound of the settling-time function. Numerical experiments here validate these advanced features further, emphasizing the efficacy and excellent performance of the proposed PIP-ZNN model.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"13 4","pages":"1565-1576"},"PeriodicalIF":5.4,"publicationDate":"2025-11-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145729402","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 : 2025-11-05DOI: 10.1109/TETC.2025.3626943
Muhammad Ahmad;Manuel Mazzara;Salvatore Distefano;Adil Mehmood Khan;Muhammad Hassaan Farooq Butt;Muhammad Usama;Danfeng Hong
Hyperspectral image (HSI) classification plays a pivotal role in domains such as environmental monitoring, agriculture, and urban planning. Traditional methods, including conventional machine learning and convolutional neural networks (CNNs), often struggle to effectively capture intricate spectral-spatial features and global contextual information. Transformer-based models, while powerful in capturing long-range dependencies, often demand substantial computational resources, posing challenges in scenarios where labeled datasets are limited, as in HSI applications. To overcome such challenges, this work proposes GraphMamba, a hybrid model that combines spectral-spatial token generation, graph-based token prioritization, and cross-attention mechanisms. The model introduces a novel hybridization of state-space modeling and Gated Recurrent Units (GRU), capturing both linear and nonlinear spatial-spectral dynamics. This approach enhances the ability to model complex spatial-spectral relationships while maintaining scalability and computational efficiency across diverse HSI datasets. Through comprehensive experiments, we demonstrate that GraphMamba outperforms existing state-of-the-art models, offering a scalable and robust solution for complex HSI classification tasks.
{"title":"GraphMamba: Graph Tokenization Mamba for Hyperspectral Image Classification","authors":"Muhammad Ahmad;Manuel Mazzara;Salvatore Distefano;Adil Mehmood Khan;Muhammad Hassaan Farooq Butt;Muhammad Usama;Danfeng Hong","doi":"10.1109/TETC.2025.3626943","DOIUrl":"https://doi.org/10.1109/TETC.2025.3626943","url":null,"abstract":"Hyperspectral image (HSI) classification plays a pivotal role in domains such as environmental monitoring, agriculture, and urban planning. Traditional methods, including conventional machine learning and convolutional neural networks (CNNs), often struggle to effectively capture intricate spectral-spatial features and global contextual information. Transformer-based models, while powerful in capturing long-range dependencies, often demand substantial computational resources, posing challenges in scenarios where labeled datasets are limited, as in HSI applications. To overcome such challenges, this work proposes GraphMamba, a hybrid model that combines spectral-spatial token generation, graph-based token prioritization, and cross-attention mechanisms. The model introduces a novel hybridization of state-space modeling and Gated Recurrent Units (GRU), capturing both linear and nonlinear spatial-spectral dynamics. This approach enhances the ability to model complex spatial-spectral relationships while maintaining scalability and computational efficiency across diverse HSI datasets. Through comprehensive experiments, we demonstrate that GraphMamba outperforms existing state-of-the-art models, offering a scalable and robust solution for complex HSI classification tasks.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"13 4","pages":"1510-1521"},"PeriodicalIF":5.4,"publicationDate":"2025-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145729314","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 : 2025-10-06DOI: 10.1109/TETC.2025.3615827
Josie E. Rodriguez Condia;Juan-David Guerrero-Balaguera;Robert Limas Sierra;Matteo Sonza Reorda
Mature computing formats, such as Floating-Point (FP), provide optimal accuracy to process real values and are essential in most scientific domains. However, the massive market adoption of highly parallel systems, with advanced technology nodes, in several domains exacerbates the need for highly reliable systems. Formerly, most reliability evaluations targeted FP hardware. Unfortunately, fine-grain assessments on cores with recent arithmetic format alternatives, such as Posit (particularly suited for Artificial Intelligence), have remained partially unexplored. Similarly, the effects of corruption on operations due to faulty hardware are not well-known, which may prevent the proposal of effective mitigation mechanisms. This work exhaustively evaluates the fine-grain effects of permanent faults in the hardware of arithmetic cores for the three most extensively used operations in modern applications (Add, Multiply, and Multiply and Add), including machine learning, implemented in Posit and FP. Our results indicate that Posit cores are less fault-vulnerable than FP ones. However, Posit cores are more prone to induce significant operational corruption than FP ones (5.2% to 7.5%). We also found that absolute errors in faulty FP cores are higher by up to 2 orders of magnitude than in Posit ones. Finally, we applied and evaluated three mitigation mechanisms (Self-Check and repair, Dual Modular Redundancy, and Triple Modular Redundancy), effectively reducing the most critical errors with moderate area costs (20% to 110%).
{"title":"Investigating and Mitigating Critical Faults in Floating-Point and Posit Arithmetic Hardware","authors":"Josie E. Rodriguez Condia;Juan-David Guerrero-Balaguera;Robert Limas Sierra;Matteo Sonza Reorda","doi":"10.1109/TETC.2025.3615827","DOIUrl":"https://doi.org/10.1109/TETC.2025.3615827","url":null,"abstract":"Mature computing formats, such as Floating-Point (FP), provide optimal accuracy to process real values and are essential in most scientific domains. However, the massive market adoption of highly parallel systems, with advanced technology nodes, in several domains exacerbates the need for highly reliable systems. Formerly, most reliability evaluations targeted FP hardware. Unfortunately, fine-grain assessments on cores with recent arithmetic format alternatives, such as Posit (particularly suited for Artificial Intelligence), have remained partially unexplored. Similarly, the effects of corruption on operations due to faulty hardware are not well-known, which may prevent the proposal of effective mitigation mechanisms. This work exhaustively evaluates the fine-grain effects of permanent faults in the hardware of arithmetic cores for the three most extensively used operations in modern applications (<italic>Add</i>, <italic>Multiply</i>, and <italic>Multiply and Add</i>), including machine learning, implemented in Posit and FP. Our results indicate that Posit cores are less fault-vulnerable than FP ones. However, Posit cores are more prone to induce significant operational corruption than FP ones (5.2% to 7.5%). We also found that absolute errors in faulty FP cores are higher by up to 2 orders of magnitude than in Posit ones. Finally, we applied and evaluated three mitigation mechanisms (<italic>Self-Check and repair</i>, <italic>Dual Modular Redundancy</i>, and <italic>Triple Modular Redundancy</i>), effectively reducing the most critical errors with moderate area costs (20% to 110%).","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"13 4","pages":"1605-1617"},"PeriodicalIF":5.4,"publicationDate":"2025-10-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145674752","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 : 2025-10-02DOI: 10.1109/TETC.2025.3614659
Nan Wang;Lijun Lu;Songping Liu;Hongqing Zhu;Yu Zhu
The high penetration of third-party intellectual property in MPSoCs gives rise to security concerns, and a set of security-driven constraints is imposed into the task scheduling step of the design process to protect MPSoCs against hardware Trojan attacks. Due to the significant performance and area overheads incurred, designers start to selectively apply security-driven constraints to achieve the design targets, but they often ignore that parts of a design may be more vulnerable to hardware Trojan attacks. In this study, the differences in vulnerability to hardware Trojan attacks are also considered in the MPSoC design process, and a security-driven task scheduling method is proposed to minimize both the design vulnerability and chip area under deadline constraints. First, the schedule length is iteratively optimized by a maximum weight independent set-based method that minimizes the vulnerability increment. Second, tasks are assigned to IP vendors with a minimized number of cores required by maximizing the core sharing of tasks. Finally, tasks are scheduled to time periods using the force-directed scheduling method. Experimental results demonstrate the effectiveness of the proposed method in reducing the number of cores while maintaining system security under deadline constraints.
{"title":"Security-Driven Task Scheduling Under Deadline Constraints for MPSoCs With Untrusted 3PIP Cores","authors":"Nan Wang;Lijun Lu;Songping Liu;Hongqing Zhu;Yu Zhu","doi":"10.1109/TETC.2025.3614659","DOIUrl":"https://doi.org/10.1109/TETC.2025.3614659","url":null,"abstract":"The high penetration of third-party intellectual property in MPSoCs gives rise to security concerns, and a set of security-driven constraints is imposed into the task scheduling step of the design process to protect MPSoCs against hardware Trojan attacks. Due to the significant performance and area overheads incurred, designers start to selectively apply security-driven constraints to achieve the design targets, but they often ignore that parts of a design may be more vulnerable to hardware Trojan attacks. In this study, the differences in vulnerability to hardware Trojan attacks are also considered in the MPSoC design process, and a security-driven task scheduling method is proposed to minimize both the design vulnerability and chip area under deadline constraints. First, the schedule length is iteratively optimized by a maximum weight independent set-based method that minimizes the vulnerability increment. Second, tasks are assigned to IP vendors with a minimized number of cores required by maximizing the core sharing of tasks. Finally, tasks are scheduled to time periods using the force-directed scheduling method. Experimental results demonstrate the effectiveness of the proposed method in reducing the number of cores while maintaining system security under deadline constraints.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"13 4","pages":"1577-1590"},"PeriodicalIF":5.4,"publicationDate":"2025-10-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145674811","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 : 2025-09-22DOI: 10.1109/TETC.2025.3609004
Tien-Dung Cao;Nguyen T. Vuong;Thai Q. Le;Hoang V. N. Dao;Tram Truong-Huu
In federated learning, the models can be trained synchronously or asynchronously. Many existing works have focused on developing an aggregation method for the server to aggregate multiple local models into the global model with improved performance. They ignore the heterogeneity of the training workers, which causes the delay in the training of the local models, leading to the obsolete information issue. In this paper, we design and develop Asyn2F, an Asynchronous Federated learning Framework with bidirectional model aggregation. By bidirectional aggregation, Asyn2F, on one hand, allows the server to asynchronously aggregate multiple local models and generate a new global model. On the other hand, it allows the training workers to aggregate the new version of the global model into a local model, which is being optimized even in the middle of a training epoch. We develop Asyn2F considering various practical implementation requirements with geographically distributed and heterogeneous training workers. Extensive experiments with different datasets show that the models trained by Asyn2F achieve higher performance compared to the state-of-the-art techniques. The experiments also demonstrate the effectiveness, practicality, and scalability of Asyn2F, making it ready for practical deployment.
{"title":"Asyn2F: An Asynchronous Federated Learning Framework With Bidirectional Model Aggregation","authors":"Tien-Dung Cao;Nguyen T. Vuong;Thai Q. Le;Hoang V. N. Dao;Tram Truong-Huu","doi":"10.1109/TETC.2025.3609004","DOIUrl":"https://doi.org/10.1109/TETC.2025.3609004","url":null,"abstract":"In federated learning, the models can be trained synchronously or asynchronously. Many existing works have focused on developing an aggregation method for the server to aggregate multiple local models into the global model with improved performance. They ignore the heterogeneity of the training workers, which causes the delay in the training of the local models, leading to the obsolete information issue. In this paper, we design and develop <sc>Asyn2F</small>, an <sc>Asyn</small>chronous <sc>F</small>ederated learning <sc>F</small>ramework with bidirectional model aggregation. By bidirectional aggregation, <sc>Asyn2F</small>, on one hand, allows the server to asynchronously aggregate multiple local models and generate a new global model. On the other hand, it allows the training workers to aggregate the new version of the global model into a local model, which is being optimized even in the middle of a training epoch. We develop <sc>Asyn2F</small> considering various practical implementation requirements with geographically distributed and heterogeneous training workers. Extensive experiments with different datasets show that the models trained by <sc>Asyn2F</small> achieve higher performance compared to the state-of-the-art techniques. The experiments also demonstrate the effectiveness, practicality, and scalability of <sc>Asyn2F</small>, making it ready for practical deployment.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"13 4","pages":"1618-1632"},"PeriodicalIF":5.4,"publicationDate":"2025-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145729449","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 : 2025-09-19DOI: 10.1109/TETC.2025.3609050
Alberto Bosio;Samuele Germiniani;Graziano Pravadelli;Marcello Traiola
In Approximate Computing (AxC), design exploration methods have been introduced to automatically identify approximation targets at the gate level. However, only some of them are applicable at Register Transfer Level (RTL); furthermore, the benefits of combining information from assertions and fault analysis have not been fully explored. This paper proposes a novel methodology for guiding AxC design exploration at RTL considering two approximation techniques: bit-width reduction and statement reduction. Then, it employs fault injection to mimic the approximation effect on the design under approximation. To guide the designer while assessing the approximation choices, assertions, which formally capture the behaviors implemented in the design, are dynamically generated from the RTL simulation traces. Then, the impact of fault injections on the truth values of the assertions is employed as a proxy for measuring the functional accuracy of the corresponding approximations. Based on this evaluation, a genetic algorithm is finally used to rank and cluster the approximation targets, thus providing the designer with an efficient and effective way to automatically analyze AxC variants in terms of the trade-off between accuracy and performance. The experiments carried out on state-of-the-art benchmarks show that the proposed approach represents a promising solution for the automation of AxC design exploration at RTL.
{"title":"A Genetic Approach for Automatic AxC Design Exploration at RTL Based on Assertion Mining and Fault Analysis","authors":"Alberto Bosio;Samuele Germiniani;Graziano Pravadelli;Marcello Traiola","doi":"10.1109/TETC.2025.3609050","DOIUrl":"https://doi.org/10.1109/TETC.2025.3609050","url":null,"abstract":"In Approximate Computing (AxC), design exploration methods have been introduced to automatically identify approximation targets at the gate level. However, only some of them are applicable at Register Transfer Level (RTL); furthermore, the benefits of combining information from assertions and fault analysis have not been fully explored. This paper proposes a novel methodology for guiding AxC design exploration at RTL considering two approximation techniques: bit-width reduction and statement reduction. Then, it employs fault injection to mimic the approximation effect on the design under approximation. To guide the designer while assessing the approximation choices, assertions, which formally capture the behaviors implemented in the design, are dynamically generated from the RTL simulation traces. Then, the impact of fault injections on the truth values of the assertions is employed as a proxy for measuring the functional accuracy of the corresponding approximations. Based on this evaluation, a genetic algorithm is finally used to rank and cluster the approximation targets, thus providing the designer with an efficient and effective way to automatically analyze AxC variants in terms of the trade-off between accuracy and performance. The experiments carried out on state-of-the-art benchmarks show that the proposed approach represents a promising solution for the automation of AxC design exploration at RTL.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"13 4","pages":"1633-1648"},"PeriodicalIF":5.4,"publicationDate":"2025-09-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11174094","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145729312","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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}
Episode mining is an active subfield of data mining in which the aim is to retrieve important knowledge from temporal data and can be used to analyze fault reports and web navigation logs. However, existing methods generally do not consider time gap constraints, and overestimate the frequency of episodes, which may lead to mining a large number of episodes that users are not interested in. To tackle this problem, this paper investigates one-off episode rule (OER) mining with time gap constraints for process event logs and proposes a one-off episode rule mining algorithm called OER-Miner that can mine frequent one-off episodes and the implicit relationship among them. To generate fewer and prune unpromising candidate episodes, OER-Miner utilizes episode join and pruning strategies, respectively. To efficiently calculate the candidate episode support, position indexes, and depth-first search and backtracking strategies are applied to calculate the number of occurrences. Experimental results verify that OER-Miner yields a better performance than seven other competitive algorithms on nine publicly available event logs. More importantly, OER-Miner can be applied to a real-industrial log to identify rework phenomena in the production process by mining strong one-off episode rules, to discover the optimal processes and deficiencies of the system, and provide recommendations for further improvement.
{"title":"OER-Miner: One-Off Episode Rule Mining for Process Event Logs","authors":"Youxi Wu;Zhihong Dong;Jing Liu;Yan Li;Cong Liu;Lijie Wen;Xindong Wu","doi":"10.1109/TETC.2025.3607892","DOIUrl":"https://doi.org/10.1109/TETC.2025.3607892","url":null,"abstract":"Episode mining is an active subfield of data mining in which the aim is to retrieve important knowledge from temporal data and can be used to analyze fault reports and web navigation logs. However, existing methods generally do not consider time gap constraints, and overestimate the frequency of episodes, which may lead to mining a large number of episodes that users are not interested in. To tackle this problem, this paper investigates one-off episode rule (OER) mining with time gap constraints for process event logs and proposes a one-off episode rule mining algorithm called OER-Miner that can mine frequent one-off episodes and the implicit relationship among them. To generate fewer and prune unpromising candidate episodes, OER-Miner utilizes episode join and pruning strategies, respectively. To efficiently calculate the candidate episode support, position indexes, and depth-first search and backtracking strategies are applied to calculate the number of occurrences. Experimental results verify that OER-Miner yields a better performance than seven other competitive algorithms on nine publicly available event logs. More importantly, OER-Miner can be applied to a real-industrial log to identify rework phenomena in the production process by mining strong one-off episode rules, to discover the optimal processes and deficiencies of the system, and provide recommendations for further improvement.","PeriodicalId":13156,"journal":{"name":"IEEE Transactions on Emerging Topics in Computing","volume":"13 4","pages":"1497-1509"},"PeriodicalIF":5.4,"publicationDate":"2025-09-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145674812","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：