Pervasive and Mobile Computing最新文献

FreTransLS: Frequency Transformer based large-scale group activity recognition model for sensor data FreTransLS：基于变频器的传感器数据大规模群体活动识别模型

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2026-01-28 DOI: 10.1016/j.pmcj.2026.102179

Ruohong Huan, Meijiao Cao, Yantong Zhou, Ji Zhang, Peng Chen, Guodao Sun, Ronghua Liang

In large-scale group activities, participants engage in a wider variety of actions, and the interactions among them become significantly more complex. This gives rise to challenges including synchronization and coordination analysis in group activity recognition. As a result, existing methods designed for recognizing small-scale group activities using sensor data often lead to inaccurate identification of dynamic patterns in large-scale settings. To address this issue, this paper proposes FreTransLS—a frequency Transformer-based model for large-scale group activity recognition using sensor data. FreTransLS introduces a novel approach for extracting time–frequency features in large-scale group activities. The approach integrates a spatio-temporal graph convolutional network (ST-GCN) module to capture spatio-temporal features within the group, along with a group location feature extraction (GLFE) module to acquire group location features. These two feature streams are fused to derive comprehensive time-domain representations of group activities. Furthermore, FreTransLS incorporates a frequency Transformer encoder built around a frequency attention mechanism. This encoder performs global analysis in the frequency domain to better model synchronization and coordination patterns in group activities. To enhance the generalization capability of the model, FreTransLS adopts a joint optimization strategy through complementary classification and reconstruction modules, which jointly refine the extracted time–frequency features. Experiments on two public datasets demonstrate that the proposed method effectively captures discriminative features from sensor data in large-scale group scenarios, leading to improved accuracy and robustness in group activity recognition.

在大规模的群体活动中，参与者参与的行动种类越来越多，他们之间的互动也变得更加复杂。这给群体活动识别中的同步性和协调性分析带来了挑战。因此，现有的利用传感器数据识别小规模群体活动的方法往往无法准确识别大规模环境下的动态模式。为了解决这个问题，本文提出了fretransls -一种基于频率转换器的模型，用于使用传感器数据进行大规模群体活动识别。FreTransLS提出了一种新的大规模群体活动时频特征提取方法。该方法集成了一个时空图卷积网络（ST-GCN）模块来捕获群体内的时空特征，以及一个群体位置特征提取（GLFE）模块来获取群体位置特征。这两种特征流融合在一起，得到了群体活动的综合时域表示。此外，FreTransLS集成了一个围绕频率注意机制构建的频率转换器编码器。该编码器在频域执行全局分析，以更好地模拟群体活动中的同步和协调模式。为了增强模型的泛化能力，FreTransLS采用互补分类和重构模块的联合优化策略，共同细化提取的时频特征。在两个公共数据集上的实验表明，该方法能有效地捕获大规模群体场景下传感器数据的判别特征，提高了群体活动识别的准确性和鲁棒性。

{"title":"FreTransLS: Frequency Transformer based large-scale group activity recognition model for sensor data","authors":"Ruohong Huan, Meijiao Cao, Yantong Zhou, Ji Zhang, Peng Chen, Guodao Sun, Ronghua Liang","doi":"10.1016/j.pmcj.2026.102179","DOIUrl":"10.1016/j.pmcj.2026.102179","url":null,"abstract":"<div><div>In large-scale group activities, participants engage in a wider variety of actions, and the interactions among them become significantly more complex. This gives rise to challenges including synchronization and coordination analysis in group activity recognition. As a result, existing methods designed for recognizing small-scale group activities using sensor data often lead to inaccurate identification of dynamic patterns in large-scale settings. To address this issue, this paper proposes FreTransLS—a frequency Transformer-based model for large-scale group activity recognition using sensor data. FreTransLS introduces a novel approach for extracting time–frequency features in large-scale group activities. The approach integrates a spatio-temporal graph convolutional network (ST-GCN) module to capture spatio-temporal features within the group, along with a group location feature extraction (GLFE) module to acquire group location features. These two feature streams are fused to derive comprehensive time-domain representations of group activities. Furthermore, FreTransLS incorporates a frequency Transformer encoder built around a frequency attention mechanism. This encoder performs global analysis in the frequency domain to better model synchronization and coordination patterns in group activities. To enhance the generalization capability of the model, FreTransLS adopts a joint optimization strategy through complementary classification and reconstruction modules, which jointly refine the extracted time–frequency features. Experiments on two public datasets demonstrate that the proposed method effectively captures discriminative features from sensor data in large-scale group scenarios, leading to improved accuracy and robustness in group activity recognition.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"117 ","pages":"Article 102179"},"PeriodicalIF":3.5,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146079111","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Design and implementation of a platform for stateful agents at the edge 边缘有状态代理平台的设计和实现

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2026-01-28 DOI: 10.1016/j.pmcj.2026.102175

Claudio Cicconetti , Emanuele Carlini , Chen Chen , Roman Kolcun , Richard Mortier

Edge–cloud computing infrastructures are increasingly widespread as they combine the flexibility of cloud-native development tools with the performance and security of distributed computing environments. Function-as-a-Service has emerged as a powerful abstraction that overcomes the limitations of a micro-service architecture. However, it generally does not support stateful functions, making it unsuitable for many practical applications in, e.g., Internet of Things (IoT) and real-time analytics. In this paper, we explore a novel paradigm, based on stateful asynchronous agents, that goes beyond traditional serverless computing. We focus on several key technical aspects: programming model, deployment procedures, design of a flexible compute node, and state management. We illustrate our paradigm using the EDGELESS platform as a concrete implementation of this stateful agents’ pattern. We report proof-of-concept experiment results obtained in a testbed with heterogeneous resource-constrained edge nodes that showcase some distinguishing features of our platform: scalable management of lightweight function instances, the advantage of keeping the state local at function instances, and delegated orchestration to enable a third-party agent to make migration decisions in a group of local nodes.

边缘云计算基础设施越来越广泛，因为它们将云原生开发工具的灵活性与分布式计算环境的性能和安全性相结合。功能即服务已经作为一种强大的抽象出现，它克服了微服务架构的局限性。然而，它通常不支持有状态功能，使其不适合许多实际应用，例如物联网（IoT）和实时分析。在本文中，我们探索了一种基于有状态异步代理的新范式，它超越了传统的无服务器计算。我们关注几个关键的技术方面：编程模型、部署过程、灵活计算节点的设计和状态管理。我们使用无边界平台作为这个有状态代理模式的具体实现来说明我们的范例。我们报告了在具有异构资源约束边缘节点的测试平台中获得的概念验证实验结果，这些结果展示了我们平台的一些显着特性：轻量级功能实例的可扩展管理，在功能实例中保持状态本地的优势，以及委托编排以使第三方代理能够在一组本地节点中做出迁移决策。

{"title":"Design and implementation of a platform for stateful agents at the edge","authors":"Claudio Cicconetti , Emanuele Carlini , Chen Chen , Roman Kolcun , Richard Mortier","doi":"10.1016/j.pmcj.2026.102175","DOIUrl":"10.1016/j.pmcj.2026.102175","url":null,"abstract":"<div><div>Edge–cloud computing infrastructures are increasingly widespread as they combine the flexibility of cloud-native development tools with the performance and security of distributed computing environments. Function-as-a-Service has emerged as a powerful abstraction that overcomes the limitations of a micro-service architecture. However, it generally does not support stateful functions, making it unsuitable for many practical applications in, e.g., Internet of Things (IoT) and real-time analytics. In this paper, we explore a novel paradigm, based on stateful asynchronous agents, that goes beyond traditional serverless computing. We focus on several key technical aspects: programming model, deployment procedures, design of a flexible compute node, and state management. We illustrate our paradigm using the <em>EDGELESS</em> platform as a concrete implementation of this stateful agents’ pattern. We report proof-of-concept experiment results obtained in a testbed with heterogeneous resource-constrained edge nodes that showcase some distinguishing features of our platform: scalable management of lightweight function instances, the advantage of keeping the state local at function instances, and delegated orchestration to enable a third-party agent to make migration decisions in a group of local nodes.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"117 ","pages":"Article 102175"},"PeriodicalIF":3.5,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146079113","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Delay optimized task offloading and performance evaluation in Fog-Enabled IoT networks 基于雾的物联网网络延迟优化任务卸载和性能评估

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2026-01-23 DOI: 10.1016/j.pmcj.2026.102177

Megha Sharma , Abhinav Tomar , Abhishek Hazra

The rapid growth and proliferation of Internet of Things (IoT) applications have intensified the demand for low-latency, energy-efficient task processing at the network edge. Fog computing has emerged as a key enabler to address these challenges by offloading computational workloads from resource-constrained sensor nodes to nearby fog nodes. In this context, we propose TSTO, a Thompson Sampling-based task offloading framework tailored for dynamic and non-stationary fog-enabled IoT environments. The scheme employs a two-tier mechanism: a greedy utility-based fog node selection followed by probabilistic decision-making using Thompson Sampling, ensuring balanced exploration and exploitation in volatile network states. To accelerate learning, a precomputation module estimates early rewards for tasks with optimistic deadlines. We provide a comprehensive delay-aware mathematical formulation, analyze the time complexity of the algorithm, and validate its scalability. Simulation results demonstrate that TSTO outperforms baseline methods such as D2CIT and BLOT, achieving up to 6% lower latency and 5% improved energy efficiency. Additionally, prototype-level validation using Raspberry Pi devices highlights the real-world applicability of the proposed model. These results confirm TSTO’s suitability for adaptive and intelligent task offloading in next-generation fog-assisted IoT systems.

物联网（IoT）应用的快速增长和扩散加剧了对网络边缘低延迟、节能任务处理的需求。通过将计算工作负载从资源受限的传感器节点卸载到附近的雾节点，雾计算已经成为解决这些挑战的关键推动者。在这种情况下，我们提出了TSTO，这是一种基于汤普森采样的任务卸载框架，专为动态和非静态雾支持物联网环境量身定制。该方案采用两层机制：基于贪婪效用的雾节点选择，然后使用汤普森采样进行概率决策，确保在不稳定的网络状态下平衡探索和利用。为了加速学习，预计算模块估计了具有乐观截止日期的任务的早期奖励。我们提供了一个全面的延迟感知数学公式，分析了算法的时间复杂度，并验证了其可扩展性。仿真结果表明，TSTO优于D2CIT和BLOT等基准方法，延迟降低6%，能效提高5%。此外，使用树莓派设备的原型级验证突出了所提出模型的实际适用性。这些结果证实了TSTO在下一代雾辅助物联网系统中的适应性和智能任务卸载的适用性。

{"title":"Delay optimized task offloading and performance evaluation in Fog-Enabled IoT networks","authors":"Megha Sharma , Abhinav Tomar , Abhishek Hazra","doi":"10.1016/j.pmcj.2026.102177","DOIUrl":"10.1016/j.pmcj.2026.102177","url":null,"abstract":"<div><div>The rapid growth and proliferation of Internet of Things (IoT) applications have intensified the demand for low-latency, energy-efficient task processing at the network edge. Fog computing has emerged as a key enabler to address these challenges by offloading computational workloads from resource-constrained sensor nodes to nearby fog nodes. In this context, we propose TSTO, a Thompson Sampling-based task offloading framework tailored for dynamic and non-stationary fog-enabled IoT environments. The scheme employs a two-tier mechanism: a greedy utility-based fog node selection followed by probabilistic decision-making using Thompson Sampling, ensuring balanced exploration and exploitation in volatile network states. To accelerate learning, a precomputation module estimates early rewards for tasks with optimistic deadlines. We provide a comprehensive delay-aware mathematical formulation, analyze the time complexity of the algorithm, and validate its scalability. Simulation results demonstrate that TSTO outperforms baseline methods such as D2CIT and BLOT, achieving up to 6% lower latency and 5% improved energy efficiency. Additionally, prototype-level validation using Raspberry Pi devices highlights the real-world applicability of the proposed model. These results confirm TSTO’s suitability for adaptive and intelligent task offloading in next-generation fog-assisted IoT systems.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"117 ","pages":"Article 102177"},"PeriodicalIF":3.5,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146024310","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Centralized sequential federated learning: single-server simulation for cross-region load forecasting 集中式顺序联邦学习：跨区域负载预测的单服务器模拟

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2026-01-23 DOI: 10.1016/j.pmcj.2026.102176

Cong Zhou , Ming Li , Longfa Yuan , Nanwei Ding , Ruijun Tie , Zheng Zeng , Zhen Li

To address data distribution heterogeneity (non-IID) and training-efficiency challenges in cross-regional electric load forecasting, this paper proposes a Centralized Sequential Federated Learning (CSFL) framework. In a single-server, centrally stored setting, CSFL emulates multi-regional “clients” via logical partitions, employs sequential local training with central aggregation, and incorporates a dynamic learning-rate decay coupled with round resets to promote progressive integration of cross-regional features. Combined with a feature-engineering pipeline based on Ward’s minimum-variance clustering and Pearson correlation analysis, CSFL substantially improves cross-regional forecasting performance. Experiments show that, compared with directly applying a single local model across regions, CSFL reduces the average forecasting error by 22.3%, and its gains are statistically significant according to five independent runs with a paired t-test (

p = 0.032

). The method achieves efficient cross-regional knowledge fusion in a single-server environment, offering a high-performing and easily deployable solution for power grid dispatch centers.

为了解决跨区域电力负荷预测中的数据分布异质性和训练效率问题，本文提出了一种集中式顺序联邦学习（CSFL）框架。在单服务器、集中存储设置中，CSFL通过逻辑分区模拟多区域的“客户端”，采用具有中央聚合的顺序局部训练，并结合动态学习率衰减与轮重置相结合，以促进跨区域特征的逐步集成。结合基于Ward最小方差聚类和Pearson相关分析的特征工程管道，CSFL大大提高了跨区域预测性能。实验表明，与直接跨区域应用单一局部模型相比，CSFL平均预测误差降低了22.3%，经5次独立运行的配对t检验，其收益具有统计学意义（p=0.032）。该方法在单服务器环境下实现了高效的跨区域知识融合，为电网调度中心提供了一种高性能、易部署的解决方案。

{"title":"Centralized sequential federated learning: single-server simulation for cross-region load forecasting","authors":"Cong Zhou , Ming Li , Longfa Yuan , Nanwei Ding , Ruijun Tie , Zheng Zeng , Zhen Li","doi":"10.1016/j.pmcj.2026.102176","DOIUrl":"10.1016/j.pmcj.2026.102176","url":null,"abstract":"<div><div>To address data distribution heterogeneity (non-IID) and training-efficiency challenges in cross-regional electric load forecasting, this paper proposes a Centralized Sequential Federated Learning (CSFL) framework. In a single-server, centrally stored setting, CSFL emulates multi-regional “clients” via logical partitions, employs sequential local training with central aggregation, and incorporates a dynamic learning-rate decay coupled with round resets to promote progressive integration of cross-regional features. Combined with a feature-engineering pipeline based on Ward’s minimum-variance clustering and Pearson correlation analysis, CSFL substantially improves cross-regional forecasting performance. Experiments show that, compared with directly applying a single local model across regions, CSFL reduces the average forecasting error by 22.3%, and its gains are statistically significant according to five independent runs with a paired <em>t</em>-test (<span><math><mrow><mi>p</mi><mo>=</mo><mn>0.032</mn></mrow></math></span>). The method achieves efficient cross-regional knowledge fusion in a single-server environment, offering a high-performing and easily deployable solution for power grid dispatch centers.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"117 ","pages":"Article 102176"},"PeriodicalIF":3.5,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146079114","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

ALIBIS: Assessing and mitigating the risk of sensitive metadata Leakage In moBile Image Sharing ALIBIS：移动图像共享中敏感元数据泄露的评估和降低风险

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2026-01-17 DOI: 10.1016/j.pmcj.2026.102171

Tran Thanh Lam Nguyen, Barbara Carminati, Elena Ferrari

Smartphones have become necessary in modern life and can replace traditional devices like cameras. The high demand for taking and sharing photos via smartphones, especially with the explosion of social networks and instant messaging, highlights the importance of smartphones. Android, the leading smartphone operating system, has continuously improved user security and privacy over its 17 years of development (2008–2025). However, security vulnerabilities still exist because of its open-source nature. This paper introduces ALIBIS, a framework that automatically estimates the risk of leakage of sensitive data contained in EXIF metadata when users share images online by combining static analysis and Large Language Models (LLMs). ALIBIS demonstrates consistent and robust estimation capabilities, achieving an average accuracy, precision, recall, and f1 score in k-fold cross-validation (k=5) of 0.8686, 0.8902, 0.881, and 0.8854, respectively. In addition, a survey of 130 global participants, including Android app developers and end-users, revealed a significant lack of awareness about image metadata and its risks: 82.3% of participants (user role) do not delete sensitive metadata before sharing images, and 62.3% do not know how to remove metadata. Furthermore, only 1.9% of participants (developer role) proactively remove EXIF metadata during programming. We propose ExifMetadataLib, a lightweight library for easy integration with Android OS, to mitigate sensitive metadata leakage.

智能手机已经成为现代生活的必需品，可以取代相机等传统设备。通过智能手机拍摄和分享照片的高需求，尤其是随着社交网络和即时通讯的爆炸式增长，凸显了智能手机的重要性。Android作为领先的智能手机操作系统，在其17年的发展（2008-2025）中不断提高用户的安全性和隐私性。然而，由于其开源特性，安全漏洞仍然存在。本文介绍了ALIBIS框架，该框架结合静态分析和大型语言模型（Large Language Models, llm），自动估计用户在线共享图像时EXIF元数据中包含的敏感数据泄露的风险。ALIBIS具有一致性和鲁棒性的估计能力，k-fold交叉验证（k=5）的平均正确率、精密度、召回率和f1得分分别为0.8686、0.8902、0.881和0.8854。此外，一项针对130名全球参与者（包括Android应用开发者和最终用户）的调查显示，人们对图像元数据及其风险的认识严重不足：82.3%的参与者（用户角色）在共享图像之前不会删除敏感元数据，62.3%的参与者不知道如何删除元数据。此外，只有1.9%的参与者（开发人员角色）在编程期间主动删除EXIF元数据。我们提出ExifMetadataLib，一个轻量级的库，易于与Android操作系统集成，以减轻敏感的元数据泄漏。

{"title":"ALIBIS: Assessing and mitigating the risk of sensitive metadata Leakage In moBile Image Sharing","authors":"Tran Thanh Lam Nguyen, Barbara Carminati, Elena Ferrari","doi":"10.1016/j.pmcj.2026.102171","DOIUrl":"10.1016/j.pmcj.2026.102171","url":null,"abstract":"<div><div>Smartphones have become necessary in modern life and can replace traditional devices like cameras. The high demand for taking and sharing photos via smartphones, especially with the explosion of social networks and instant messaging, highlights the importance of smartphones. Android, the leading smartphone operating system, has continuously improved user security and privacy over its 17 years of development (2008–2025). However, security vulnerabilities still exist because of its open-source nature. This paper introduces ALIBIS, a framework that automatically estimates the risk of leakage of sensitive data contained in EXIF metadata when users share images online by combining static analysis and Large Language Models (LLMs). ALIBIS demonstrates consistent and robust estimation capabilities, achieving an average accuracy, precision, recall, and f1 score in k-fold cross-validation (k=5) of 0.8686, 0.8902, 0.881, and 0.8854, respectively. In addition, a survey of 130 global participants, including Android app developers and end-users, revealed a significant lack of awareness about image metadata and its risks: 82.3% of participants (user role) do not delete sensitive metadata before sharing images, and 62.3% do not know how to remove metadata. Furthermore, only 1.9% of participants (developer role) proactively remove EXIF metadata during programming. We propose ExifMetadataLib, a lightweight library for easy integration with Android OS, to mitigate sensitive metadata leakage.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"117 ","pages":"Article 102171"},"PeriodicalIF":3.5,"publicationDate":"2026-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146024341","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Mobility-aware Q-learning for workload offloading in vehicular edge–cloud environment 基于移动感知q学习的车辆边缘云环境下的工作负载卸载

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2026-01-13 DOI: 10.1016/j.pmcj.2026.102172

Afzal Badshah , Abdulrahman Ahmed Gharawi , Mona Eisa , Nada Alzaben , Saud Yonbawi , Ali Daud

The Intelligent Transportation System (ITS) continuously generates data that needs to be processed under strict latency and connectivity constraints across a heterogeneous computing architecture (e.g., Vehicular Edge Computing (VEC), Mobile Edge Computing (MEC), and Cloud Computing (CC)). In this context, efficient task offloading requires mobility and server-aware intelligence to optimize communication delay, cost, and resource utilization. In this paper, we propose a mobility-aware Q-learning offloading scheduler that learns optimal tier selection on real-time metrics (e.g., resource availability, signal strength, and Base Station (BS) handover dynamics). Unlike the previous investigation, this approach explicitly incorporates vehicle mobility patterns to the offloading decision using Q-learning. The scheduler favors VEC when underutilized, transitions to MEC when the VEC is overutilized, and falls back to the cloud only when VEC and MEC are infeasible. A structured reward model reinforces decisions that improve resource efficiency and penalizes excessive switching or skipping underutilized resources. The proposed framework is evaluated using DriveNetSim, a custom-developed vehicular simulator that models realistic mobility, signal degradation, and BS switching. Simulation results show a strong preference for VEC, with shifts to MEC only under VEC over-utilization and minimal reliance on the cloud. As a result, the system achieves up to 43% reduction in transmission delay and 38% reduction in processing cost, validating its effectiveness in dynamic vehicular environments.

智能交通系统（ITS）不断生成数据，这些数据需要在严格的延迟和连接约束下跨异构计算架构（例如，车辆边缘计算（VEC），移动边缘计算（MEC）和云计算（CC））进行处理。在这种情况下，高效的任务卸载需要机动性和服务器感知智能来优化通信延迟、成本和资源利用率。在本文中，我们提出了一个移动性感知的q -学习卸载调度程序，该调度程序根据实时指标（例如，资源可用性，信号强度和基站（BS）切换动态）学习最佳层选择。与之前的研究不同，该方法明确地将车辆移动模式结合到使用q学习的卸载决策中。调度程序在未充分利用时倾向于VEC，在VEC被过度利用时过渡到MEC，只有在VEC和MEC不可行的情况下才会回到云。结构化的奖励模式强化了提高资源效率的决策，并惩罚过度转换或跳过未充分利用的资源。所提出的框架使用DriveNetSim进行评估，DriveNetSim是一种定制开发的车辆模拟器，可以模拟现实的移动性、信号退化和BS切换。模拟结果显示，人们对VEC有强烈的偏好，只有在VEC过度利用和对云的依赖最小的情况下，才会转向MEC。结果表明，该系统的传输延迟降低了43%，处理成本降低了38%，验证了其在动态车辆环境中的有效性。

{"title":"Mobility-aware Q-learning for workload offloading in vehicular edge–cloud environment","authors":"Afzal Badshah , Abdulrahman Ahmed Gharawi , Mona Eisa , Nada Alzaben , Saud Yonbawi , Ali Daud","doi":"10.1016/j.pmcj.2026.102172","DOIUrl":"10.1016/j.pmcj.2026.102172","url":null,"abstract":"<div><div>The Intelligent Transportation System (ITS) continuously generates data that needs to be processed under strict latency and connectivity constraints across a heterogeneous computing architecture (e.g., Vehicular Edge Computing (VEC), Mobile Edge Computing (MEC), and Cloud Computing (CC)). In this context, efficient task offloading requires mobility and server-aware intelligence to optimize communication delay, cost, and resource utilization. In this paper, we propose a mobility-aware Q-learning offloading scheduler that learns optimal tier selection on real-time metrics (e.g., resource availability, signal strength, and Base Station (BS) handover dynamics). Unlike the previous investigation, this approach explicitly incorporates vehicle mobility patterns to the offloading decision using Q-learning. The scheduler favors VEC when underutilized, transitions to MEC when the VEC is overutilized, and falls back to the cloud only when VEC and MEC are infeasible. A structured reward model reinforces decisions that improve resource efficiency and penalizes excessive switching or skipping underutilized resources. The proposed framework is evaluated using <em>DriveNetSim</em>, a custom-developed vehicular simulator that models realistic mobility, signal degradation, and BS switching. Simulation results show a strong preference for VEC, with shifts to MEC only under VEC over-utilization and minimal reliance on the cloud. As a result, the system achieves up to 43% reduction in transmission delay and 38% reduction in processing cost, validating its effectiveness in dynamic vehicular environments.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"117 ","pages":"Article 102172"},"PeriodicalIF":3.5,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145980648","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

TCP-HAR: On-Device Transferable and Copyright-Preserving Human Activity Recognition TCP-HAR：设备上可转让和保护版权的人类活动识别

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2026-01-13 DOI: 10.1016/j.pmcj.2026.102163

Alessio Sacco , Bruno Palermo , Giulio Figliolino , Chiara Contoli , Guido Marchetto , Flavio Esposito

Teaching a machine to accurately identify human activities from sensor data poses a significant challenge, which is further compounded by considerations of data privacy, resource costs, and responsiveness, particularly within the constraints of devices like smartphones. While current solutions efficiently identify activities, trained models are barely portable in scenarios composed of diverse activities and limited battery life devices, such as smartphones. This paper introduces Transferable and Copyright-Preserving Human Activity Recognition (TCP-HAR), a mobile-based HAR system that integrates digital watermarking, Federated Learning (FL), Transfer Learning (TL), and compression techniques to provide efficient human activity recognition while providing copyright protection of deep neural network models over Android smartphones. Our solution optimizes the utilization of FL, TL, and their combination (FTL) by extensively testing standalone TL models in offline contexts and comparing these results with FL across a network of mobile devices. Our findings highlight the benefits of TCP-HAR for mobile environments in terms of accuracy, F1-score, and training time. In addition, our proposed watermarking mechanism is robust yet computationally efficient, ensuring ownership verification without compromising the scalability of the TFL process.

教机器从传感器数据中准确识别人类活动是一项重大挑战，考虑到数据隐私、资源成本和响应能力，尤其是在智能手机等设备的限制下，这一挑战进一步复杂化。虽然目前的解决方案可以有效地识别活动，但经过训练的模型在由各种活动和有限电池寿命的设备（如智能手机）组成的场景中几乎无法携带。本文介绍了可转移和保护版权的人类活动识别（TCP-HAR），这是一种基于移动的HAR系统，它集成了数字水印、联邦学习（FL）、迁移学习（TL）和压缩技术，在提供有效的人类活动识别的同时，为Android智能手机上的深度神经网络模型提供版权保护。我们的解决方案通过在离线环境中广泛测试独立的TL模型，并将这些结果与移动设备网络中的FL进行比较，优化了FL、TL及其组合（FTL）的利用率。我们的研究结果强调了TCP-HAR在准确性、f1分数和训练时间方面对移动环境的好处。此外，我们提出的水印机制鲁棒且计算效率高，确保所有权验证而不影响TFL过程的可扩展性。

{"title":"TCP-HAR: On-Device Transferable and Copyright-Preserving Human Activity Recognition","authors":"Alessio Sacco , Bruno Palermo , Giulio Figliolino , Chiara Contoli , Guido Marchetto , Flavio Esposito","doi":"10.1016/j.pmcj.2026.102163","DOIUrl":"10.1016/j.pmcj.2026.102163","url":null,"abstract":"<div><div>Teaching a machine to accurately identify human activities from sensor data poses a significant challenge, which is further compounded by considerations of data privacy, resource costs, and responsiveness, particularly within the constraints of devices like smartphones. While current solutions efficiently identify activities, trained models are barely portable in scenarios composed of diverse activities and limited battery life devices, such as smartphones. This paper introduces Transferable and Copyright-Preserving Human Activity Recognition (TCP-HAR), a mobile-based HAR system that integrates digital watermarking, Federated Learning (FL), Transfer Learning (TL), and compression techniques to provide efficient human activity recognition while providing copyright protection of deep neural network models over Android smartphones. Our solution optimizes the utilization of FL, TL, and their combination (FTL) by extensively testing standalone TL models in offline contexts and comparing these results with FL across a network of mobile devices. Our findings highlight the benefits of TCP-HAR for mobile environments in terms of accuracy, F1-score, and training time. In addition, our proposed watermarking mechanism is robust yet computationally efficient, ensuring ownership verification without compromising the scalability of the TFL process.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"117 ","pages":"Article 102163"},"PeriodicalIF":3.5,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146079112","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

SMOTE-Enhanced CNN-Bi-LSTM for wearable sensor-based human activity recognition 基于smote增强的CNN-Bi-LSTM可穿戴传感器的人体活动识别

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2026-01-08 DOI: 10.1016/j.pmcj.2026.102161

Ahmed Arafa , Hadeer Harfoush , Nawal El-Fishawy , Marwa Radad

Human Activity Recognition (HAR) refers to the automatic recognition of various human physical activities such as walking, sitting, and standing. HAR based on wearable sensors and smartphones has gained significant attention due to its wide range of applications in healthcare, sports, rehabilitation, and smart environments. Despite extensive research, challenges remain in modeling complex spatial–temporal dependencies within activity sequences and addressing class imbalance issues in sensor-based datasets. In this paper, we propose a hybrid deep learning model that integrates a Convolutional Neural Network (CNN) for spatial feature extraction followed by a Bidirectional Long Short-Term Memory (Bi-LSTM) for bi-directional sequential analysis and a fully connected layer for classifying the different types of activities. To address data imbalance and enhance the model robustness, three oversampling techniques — Random Oversampling (ROS), Adaptive Synthetic Sampling (ADASYN), and Synthetic Minority Over-sampling Technique (SMOTE) — were experimentally evaluated, with SMOTE demonstrating superior performance. The proposed model was trained and evaluated on six publicly available benchmark datasets: MHealth, PAMAP2, WISDM, UCI-HAR, USC-HAD and Opportunity datasets, achieving F1-score at 100%, 97.99%, 99.0%, 94.81%, 91.13% and 90.95% respectively. Comparative results demonstrate that the proposed framework outperforms several state-of-the-art methods across multiple datasets, confirming its robustness, reliability, and generalization capability for diverse human activity recognition scenarios.

人类活动识别（Human Activity Recognition， HAR）是指对人类行走、坐、站等各种身体活动的自动识别。基于可穿戴传感器和智能手机的HAR因其在医疗保健、运动、康复和智能环境中的广泛应用而备受关注。尽管进行了广泛的研究，但在模拟活动序列中复杂的时空依赖关系和解决基于传感器的数据集中的类不平衡问题方面仍然存在挑战。在本文中，我们提出了一种混合深度学习模型，该模型集成了用于空间特征提取的卷积神经网络（CNN）、用于双向序列分析的双向长短期记忆（Bi-LSTM）和用于分类不同类型活动的完全连接层。为了解决数据不平衡和增强模型鲁棒性，实验评估了三种过采样技术——随机过采样（ROS）、自适应合成采样（ADASYN）和合成少数过采样技术（SMOTE）， SMOTE表现出优异的性能。该模型在MHealth、PAMAP2、WISDM、UCI-HAR、USC-HAD和Opportunity 6个公开的基准数据集上进行了训练和评估，f1得分分别为100%、97.99%、99.0%、94.81%、91.13%和90.95%。对比结果表明，所提出的框架在多个数据集上优于几种最先进的方法，证实了其在不同人类活动识别场景中的鲁棒性、可靠性和泛化能力。

{"title":"SMOTE-Enhanced CNN-Bi-LSTM for wearable sensor-based human activity recognition","authors":"Ahmed Arafa , Hadeer Harfoush , Nawal El-Fishawy , Marwa Radad","doi":"10.1016/j.pmcj.2026.102161","DOIUrl":"10.1016/j.pmcj.2026.102161","url":null,"abstract":"<div><div>Human Activity Recognition (HAR) refers to the automatic recognition of various human physical activities such as walking, sitting, and standing. HAR based on wearable sensors and smartphones has gained significant attention due to its wide range of applications in healthcare, sports, rehabilitation, and smart environments. Despite extensive research, challenges remain in modeling complex spatial–temporal dependencies within activity sequences and addressing class imbalance issues in sensor-based datasets. In this paper, we propose a hybrid deep learning model that integrates a Convolutional Neural Network (CNN) for spatial feature extraction followed by a Bidirectional Long Short-Term Memory (Bi-LSTM) for bi-directional sequential analysis and a fully connected layer for classifying the different types of activities. To address data imbalance and enhance the model robustness, three oversampling techniques — Random Oversampling (ROS), Adaptive Synthetic Sampling (ADASYN), and Synthetic Minority Over-sampling Technique (SMOTE) — were experimentally evaluated, with SMOTE demonstrating superior performance. The proposed model was trained and evaluated on six publicly available benchmark datasets: MHealth, PAMAP2, WISDM, UCI-HAR, USC-HAD and Opportunity datasets, achieving F1-score at 100%, 97.99%, 99.0%, 94.81%, 91.13% and 90.95% respectively. Comparative results demonstrate that the proposed framework outperforms several state-of-the-art methods across multiple datasets, confirming its robustness, reliability, and generalization capability for diverse human activity recognition scenarios.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"117 ","pages":"Article 102161"},"PeriodicalIF":3.5,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145915147","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

MDWD-KAN: Multilevel discrete wavelet decomposition with Kolmogorov–Arnold network for fall detection and activity recognition using wearable sensors 基于Kolmogorov-Arnold网络的多电平离散小波分解可穿戴传感器跌倒检测和活动识别

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2026-01-06 DOI: 10.1016/j.pmcj.2026.102160

Zhiyuan Jiang , Sike Ni , Mohammed A.A. Al-qaness

Fall detection and Human Activity Recognition (HAR) are crucial applications in pervasive and mobile computing, enabling real-time monitoring of individuals – especially the elderly or patients – for enhanced safety and health management. Wearable devices have emerged as a critical tool for continuous activity monitoring, enabling real-time detection and intervention. However, the quality of data collected by wearable sensors faces several challenges, including noise interference, instability due to wearing, and individual differences. To address these challenges, this paper proposes a feature stepwise fusion detection system based on a multilevel discrete wavelet decomposition with Kolmogorov–Arnold Network, namely MDWD-KAN. This model utilizes multilevel wavelet decomposition to perform multiresolution analysis on sensor signals, extracting multilevel features and effectively enhancing feature stability and noise resistance. Additionally, through a heterogeneous model and a multilevel feature fusion strategy, MDWD-KAN achieves complementary low-frequency and high-frequency features, improving the recognition capability for complex motion patterns. Experiments were conducted on three public datasets: MobiAct, SisFall, and UniMiB-SHAR. The results show that MDWD-KAN achieves average recognition accuracies of 99.67%, 99.90%, and 99.65%, respectively, for binary classification (fall and non-fall), and 98.85%, 85.45%, and 96.86%, respectively, for multiclassification.

跌倒检测和人体活动识别（HAR）是普及和移动计算中的关键应用，能够实时监测个人，特别是老年人或患者，以加强安全和健康管理。可穿戴设备已经成为持续活动监测的关键工具，可以实现实时检测和干预。然而，可穿戴传感器收集的数据质量面临着一些挑战，包括噪声干扰、佩戴不稳定以及个体差异。为了解决这些问题，本文提出了一种基于Kolmogorov-Arnold网络的多层离散小波分解的特征逐步融合检测系统MDWD-KAN。该模型利用多级小波分解对传感器信号进行多分辨率分析，提取多级特征，有效增强特征稳定性和抗噪性。此外，MDWD-KAN通过异构模型和多层次特征融合策略，实现低频和高频特征的互补，提高了对复杂运动模式的识别能力。实验在三个公共数据集上进行：MobiAct、SisFall和unimib - share。结果表明，mddd - kan对二分类（跌倒和非跌倒）的平均识别准确率分别为99.67%、99.90%和99.65%，对多分类的平均识别准确率分别为98.85%、85.45%和96.86%。

{"title":"MDWD-KAN: Multilevel discrete wavelet decomposition with Kolmogorov–Arnold network for fall detection and activity recognition using wearable sensors","authors":"Zhiyuan Jiang , Sike Ni , Mohammed A.A. Al-qaness","doi":"10.1016/j.pmcj.2026.102160","DOIUrl":"10.1016/j.pmcj.2026.102160","url":null,"abstract":"<div><div>Fall detection and Human Activity Recognition (HAR) are crucial applications in pervasive and mobile computing, enabling real-time monitoring of individuals – especially the elderly or patients – for enhanced safety and health management. Wearable devices have emerged as a critical tool for continuous activity monitoring, enabling real-time detection and intervention. However, the quality of data collected by wearable sensors faces several challenges, including noise interference, instability due to wearing, and individual differences. To address these challenges, this paper proposes a feature stepwise fusion detection system based on a multilevel discrete wavelet decomposition with Kolmogorov–Arnold Network, namely MDWD-KAN. This model utilizes multilevel wavelet decomposition to perform multiresolution analysis on sensor signals, extracting multilevel features and effectively enhancing feature stability and noise resistance. Additionally, through a heterogeneous model and a multilevel feature fusion strategy, MDWD-KAN achieves complementary low-frequency and high-frequency features, improving the recognition capability for complex motion patterns. Experiments were conducted on three public datasets: MobiAct, SisFall, and UniMiB-SHAR. The results show that MDWD-KAN achieves average recognition accuracies of 99.67%, 99.90%, and 99.65%, respectively, for binary classification (fall and non-fall), and 98.85%, 85.45%, and 96.86%, respectively, for multiclassification.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"116 ","pages":"Article 102160"},"PeriodicalIF":3.5,"publicationDate":"2026-01-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145925535","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

FedEMMD: Entropy and MMD-based data and aggregation selection for non-iid and long-tailed data in federated learning FedEMMD：联邦学习中基于熵和mmd的数据和非id和长尾数据的聚合选择

IF 3.5 3区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Pervasive and Mobile Computing

Pub Date : 2025-12-30 DOI: 10.1016/j.pmcj.2025.102159

Nafas Gul Saadat, Santhosh Kumar G.

The increasing need for privacy-preserving machine learning has rendered centralized data collection progressively unfeasible. To solve this, Federated Learning (FL) has emerged as a distributed learning paradigm in which multiple clients collectively train a shared global model while keeping all data locally, ensuring that no private data is sent over the network. However, FL is often hindered by statistical heterogeneity, where clients’ data are non-independent and identically distributed (non-iid), resulting in biased local updates and reduced global model performance. To overcome these key challenges, this study proposes FedEMMD, a novel method to enhance model performance under heterogeneous data. First, entropy-based data selection is used to identify and select high-quality data with a lower degree of non-iidness. Second, Maximum Mean Discrepancy (MMD) is used to calculate the divergence between local updates and the global model, guaranteeing that only stable and consistent updates are aggregated on the global model. Experiments have been conducted in two heterogeneous settings (non-iid and long-tailed distribution), using CIFAR-10 and CIFAR-10-LT. Additionally, we conduct experiments with centralized Machine Learning (ML) under the same settings to establish a baseline to evaluate the effect of data heterogeneity on centralized ML. The experimental results demonstrate that FedEMMD outperforms state-of-the-art algorithms such as FedAvg, FedProx, Scaffold, and FedOpt in terms of accuracy and convergence speed in both non-iid and long-tailed scenarios, thereby improving robustness and performance under heterogeneous settings.

对保护隐私的机器学习的需求日益增长，使得集中收集数据变得越来越不可行。为了解决这个问题，联邦学习（FL）作为一种分布式学习范例出现了，在这种范例中，多个客户端共同训练一个共享的全局模型，同时将所有数据保存在本地，确保没有私有数据通过网络发送。然而，FL经常受到统计异质性的阻碍，其中客户端的数据是非独立和同分布的（non-iid），导致局部更新有偏差，降低了全局模型的性能。为了克服这些关键挑战，本研究提出了一种新的方法FedEMMD来提高异构数据下的模型性能。首先，使用基于熵的数据选择来识别和选择非完整性程度较低的高质量数据。其次，利用最大平均差异（Maximum Mean difference， MMD）计算局部更新与全局模型之间的差异，保证在全局模型上只聚合稳定一致的更新。使用CIFAR-10和CIFAR-10- lt在两种异质环境下（非id分布和长尾分布）进行了实验。此外，我们在相同设置下对集中式机器学习（ML）进行了实验，以建立基线来评估数据异质性对集中式机器学习的影响。实验结果表明，FedEMMD在非id和长尾场景下的准确性和收敛速度都优于FedAvg、FedProx、Scaffold和FedOpt等最先进的算法，从而提高了异构设置下的鲁棒性和性能。

{"title":"FedEMMD: Entropy and MMD-based data and aggregation selection for non-iid and long-tailed data in federated learning","authors":"Nafas Gul Saadat, Santhosh Kumar G.","doi":"10.1016/j.pmcj.2025.102159","DOIUrl":"10.1016/j.pmcj.2025.102159","url":null,"abstract":"<div><div>The increasing need for privacy-preserving machine learning has rendered centralized data collection progressively unfeasible. To solve this, Federated Learning (FL) has emerged as a distributed learning paradigm in which multiple clients collectively train a shared global model while keeping all data locally, ensuring that no private data is sent over the network. However, FL is often hindered by statistical heterogeneity, where clients’ data are non-independent and identically distributed (non-iid), resulting in biased local updates and reduced global model performance. To overcome these key challenges, this study proposes FedEMMD, a novel method to enhance model performance under heterogeneous data. First, entropy-based data selection is used to identify and select high-quality data with a lower degree of non-iidness. Second, Maximum Mean Discrepancy (MMD) is used to calculate the divergence between local updates and the global model, guaranteeing that only stable and consistent updates are aggregated on the global model. Experiments have been conducted in two heterogeneous settings (non-iid and long-tailed distribution), using CIFAR-10 and CIFAR-10-LT. Additionally, we conduct experiments with centralized Machine Learning (ML) under the same settings to establish a baseline to evaluate the effect of data heterogeneity on centralized ML. The experimental results demonstrate that FedEMMD outperforms state-of-the-art algorithms such as FedAvg, FedProx, Scaffold, and FedOpt in terms of accuracy and convergence speed in both non-iid and long-tailed scenarios, thereby improving robustness and performance under heterogeneous settings.</div></div>","PeriodicalId":49005,"journal":{"name":"Pervasive and Mobile Computing","volume":"117 ","pages":"Article 102159"},"PeriodicalIF":3.5,"publicationDate":"2025-12-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145915146","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0