检索结果-内蒙古大学图书馆

DDPG-AdaptConfig: A deep reinforcement learning framework for adaptive device selection and training configuration in heterogeneity federated learning

引用

FUTURE GENERATION COMPUTER SYSTEMS-THE INTERNATIONAL JOURNAL OF ESCIENCE 2025年 163卷

作者： Yu, Xinlei Gao, Zhipeng Xiong, Zijian Zhao, Chen Yang, Yang Beijing Univ Posts & Telecommun State Key Lab Networking & Switching Technol Beijing Peoples R China

Federated Learning (FL) is a distributed machine learning approach that protects user privacy by collaboratively training shared models across devices without sharing their raw personal data. Despite its advantages, FL faces issues of increased convergence time and decreased accuracy due to the heterogeneity of data and systems across devices. Existing methods for solving these issues using reinforcement learning often ignore the adaptive configuration of local training hyperparameters to suit varying data characteristics and system resources. Moreover, they frequently overlook the heterogeneous information contained within local model parameters. To address these problems, we propose the DDPG-AdaptConfig framework based on Deep Deterministic Policy Gradient (DDPG) for adaptive device selection and local training hyperparameters configuration in FL to speed up convergence and ensure high model accuracy. Additionally, we develop a new actor network that integrates the transformer mechanism to extract heterogeneous information from model parameters, which assists in device selection and hyperparameters configuration. Furthermore, we introduce a clustering-based aggregation strategy to accommodate heterogeneity and prevent performance declines. Experimental results show that our DDPG-AdaptConfig achieves significant improvements over existing baselines.

关键词： Federated learning Data heterogeneity System heterogeneity Reinforcement learning Deep deterministic policy gradient

来源：评论

学校读者我要写书评

暂无评论

PFDP: privacy-preserving federated distillation method for pretrained language models

引用

INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS 2025年 1-16页

作者： Chen, Chaomeng Su, Sen Beijing Univ Posts & Telecommun State Key Lab Networking & Switching Technol Beijing Peoples R China

Pre-trained language models (PLMs) have emerged as transformative tools in natural language processing. However, their training typically requires massive datasets that are distributed across multiple nodes, posing significant challenges for centralized training paradigms. Federated pre-trained language models (FL-PLMs) offer a promising solution by enabling collaborative model training without raw data. However, they suffer from significant performance degradation due to statistical heterogeneity across clients. Although knowledge distillation has been proposed to address this challenge by aggregating knowledge instead of model parameters, existing approaches remain vulnerable to privacy leakage and potential knowledge degradation during the distillation process. To overcome these limitations, we propose Privacy-preserving Federated Distillation for Pre-trained language models (PFDP), a novel framework designed to optimize both model performance and privacy preservation. PFDP introduces a Local Distillation with Differential Privacy (LoTDP) that safeguards client data privacy. Furthermore, we propose a Global Aggregation via Transfer Learning (GATeL) algorithm that improves the global model's generalization capabilities while mitigating the catastrophic forgetting phenomenon commonly associated with federated distillation. Extensive evaluation across multiple classification tasks demonstrates the effectiveness of PFDP in improving accuracy while preserving privacy.

关键词： Federated learning Pretrained language models Privacy-preserving

来源：评论

学校读者我要写书评

暂无评论

Cache-Aware Cooperative Multicast Beamforming in Dynamic Satellite-Terrestrial Networks

引用

IEEE TRANSACTIONS ON VEHICULAR technology 2025年第1期74卷 1433-1445页

作者： Yuan, Shuo Sun, Yaohua Peng, Mugen Beijing Univ Posts & Telecommun State Key Lab Networking & Switching Technol Beijing 100876 Peoples R China

With the burgeoning demand for data-intensive services, satellite-terrestrial networks (STNs) face increasing backhaul link congestion, deteriorating user quality of service (QoS), and escalating power consumption. Cache-aided STNs are acknowledged as a promising paradigm for accelerating content delivery to users and alleviating the load of backhaul links. However, the dynamic nature of low earth orbit (LEO) satellites and the complex interference among satellite beams and terrestrial base stations pose challenges in effectively managing limited edge resources. To address these issues, this paper proposes a method for dynamically scheduling caching and communication resources, aiming to reduce network costs in terms of transmission power consumption and backhaul traffic, while meeting user QoS demands and resource constraints. We formulate a mixed timescale problem to jointly optimize cache placement, LEO satellite beam direction, and cooperative multicast beamforming among satellite beams and base stations. To tackle this intricate problem, we propose a two-stage solution framework, where the primary problem is decoupled into a short-term content delivery subproblem and a long-term cache placement subproblem. The former subproblem is solved by designing an alternating optimization approach with whale optimization and successive convex approximation methods according to the cache placement state, while cache content in STNs is updated using an iterative algorithm that utilizes historical information. Simulation results demonstrate the effectiveness of our proposed algorithms, showcasing their convergence and significantly reducing transmission power consumption and backhaul traffic by up to 52%.

关键词： Array signal processing Satellite broadcasting Base stations Backhaul networks Satellites Optimization Multicast algorithms Power demand Space-air-ground integrated networks Low earth orbit satellites Beam direction control cache placement multicast beamforming satellite-terrestrial networks (STNs)

来源：评论

学校读者我要写书评

暂无评论

TextDiff: Enhancing scene text image super-resolution with mask-guided residual diffusion models

引用

PATTERN RECOGNITION 2025年 164卷

作者： Liu, Baolin Yang, Zongyuan Chiu, Chinwai Xiong, Yongping Beijing Univ Post & Telecommun State Key Lab Switching & Networking Technol Beijing 100876 Peoples R China

The goal of scene text image super-resolution (STISR) is to reconstruct high-resolution text-line images from unrecognizable low-resolution inputs. The existing methods relying on the optimization of pixel-level loss tend to yield text edges that exhibit a notable degree of blurring, thereby exerting a substantial impact on both the readability and recognizability of the text. To address these issues, we propose TextDiff, the first diffusion-based framework tailored for STISR. It contains two modules: the Text Enhancement Module (TEM) and the Mask- Guided Residual Diffusion Module (MRD). The TEM generates an initial deblurred text image and a mask that encodes the spatial location of the text. The MRD is responsible for effectively sharpening the text edge by modeling the residuals between the ground-truth images and the initial deblurred images. Extensive experiments demonstrate that our TextDiff achieves state-of-the-art (SOTA) performance on public benchmark datasets, with a maximum improvement of 2.0% in recognition accuracy over existing methods while enhancing the readability of scene text images. Moreover, our proposed MRD module is plug-and-play that effectively sharpens the text edges produced by SOTA methods. This enhancement not only improves the readability and recognizability of the results generated by SOTA methods but also does not require any additional joint training.

关键词： Scene text image super-resolution Text enhancement Diffusion model Multi-stage learning Model expandability

来源：评论

学校读者我要写书评

暂无评论

Hybrid Channel Tracking for THz Massive MIMO Communication Systems in Dynamic Environments

引用

IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS 2025年第3期43卷 988-1003页

作者： Fan, Yuheng Yang, Chuang Sun, Yanran Peng, Mugen Beijing Univ Posts & Telecommun State Key Lab Networking & Switching Technol Beijing 100876 Peoples R China

With gigahertz-level bandwidth, terahertz (THz) holds promise for achieving exceptionally high transmission rates in prospective sixth-generation (6G) communications. However, considerable path loss poses an obstacle to THz communications. To compensate for this, massive multiple-input-multiple-output (MIMO) based beamforming is utilized to promote directional power with narrow beams in communications. In dynamic environments, the frequent adjustment of narrow beams results in fast time-varying channel state information (CSI), which constrains the application of the THz communication systems. While traditional deterministic-based and statistical-based channel tracking methods address different aspects of this issue, they suffer from balancing accuracy and complexity in the THz dynamic environments. To solve this problem, based on the cluster distribution of THz time-varying channel, we propose a novel hybrid channel tracking method that uses deterministic physical motion variation law to extract the cluster subspace, and then statistical Markov evolution models are applied within it. To achieve this, an integrated clustering and estimation method, clustering subspace matching pursuit (CSMP) is proposed for obtaining the channel clusters prior knowledge. Then based on above hybrid tracking method design, we propose a virtual cluster subspace turbo-approximate message passing (VCS-TAMP). Finally, several simulation results validate that our proposal achieves great improvement in both accuracy and computational time performance.

关键词： Terahertz communications Vectors Massive MIMO Accuracy Heuristic algorithms Estimation Tracking Solid modeling Mathematical models Computational modeling Terahertz massive MIMO channel tracking

来源：评论

学校读者我要写书评

暂无评论

Self-Sustainable Reconfigurable Intelligent Surface-Empowered D2D Communication Network

IEEE TRANSACTIONS ON SUSTAINABLE COMPUTING

引用

IEEE TRANSACTIONS ON SUSTAINABLE COMPUTING 2025年第2期10卷 287-299页

作者： Yang, Zhixiang Feng, Lei Zhou, Fanqin Xie, Kunyi Qiu, Xuesong Li, Wenjing Beijing Univ Posts & Telecommun State Key Lab Networking & Switching Technol Beijing 100876 Peoples R China

The reconfigurable intelligent surface (RIS) is a green and promising technology that provides passive beamforming through a large amount of low-power reflecting elements, to realizes expected coverage extension and interference signal suppression. In this paper, we investigate a self-sustainable RIS-empowered D2D communication network, where the RIS first harvests energy from the D2D signals, and then uses energy collected to sustain its passive beamforming operation. We aim to characterize the energy efficiency (EE) maximization under imperfect channel state information conditions by jointly optimizing the transmit precoding in both two stages, RIS passive beamforming design, and energy harvesting time allocation. An efficient alternating optimization algorithm is proposed to deal with the difficult non-convex optimization problem. Specifically, transmit precoding is optimized by using the Dinkelbach's method, Lagrangian dual transform, quadratic transform and S-procedure. The penalty convex-concave procedure is adopted to solve the optimal phase shift of RIS. A closed-form expression for the optimal energy harvesting duration is derived. The simulation results show that the proposed scheme further enhances the EE compared with the active RIS and no RIS schemes in various scenarios.

关键词： Device-to-device communication Reconfigurable intelligent surfaces Energy harvesting Array signal processing Resource management Interference Green computing D2D communication energy efficiency maximization reconfigurable intelligent surface self-sustainable

来源：评论

学校读者我要写书评

暂无评论

Diffusion Model Empowered Efficient Data Distillation Method for Cloud-Edge Collaboration

引用

IEEE TRANSACTIONS ON COGNITIVE COMMUNICATIONS AND networking 2025年第2期11卷 902-913页

作者： Chai, Ze Lin, Yijing Gao, Zhipeng Yu, Xinlei Xie, Zhiqiang Beijing Univ Posts & Telecommun State Key Lab Networking & Switching Technol Beijing 100876 Peoples R China

The application of AI-generated models demands substantial amounts of data, which not only increases training time and memory consumption but also poses challenges to computation and communication loads in cloud-edge collaborative environments. Especially when edge devices have limited resources and network bandwidth is constrained, efficiently handling large-scale data in cloud-edge collaboration becomes a critical issue. Data distillation compresses large datasets into smaller, synthetic datasets, reducing transmission and computation overhead while maintaining model performance. However, existing data distillation methods face two main challenges: 1) improving accuracy without compromising distillation efficiency, and 2) the distilled dataset may retain backdoor attack triggers. To address these challenges, we propose an efficient data distillation method based on diffusion models, tailored for cloud-edge collaborative environments. We design a diffusion model training mechanism based on Kullback-Leibler (KL) divergence and contrastive loss to effectively distill synthetic datasets and enhance their accuracy, thereby reducing the data storage requirements on edge devices and alleviating communication burden with the cloud. Additionally, we introduce a dynamic channel weighting method based on an adaptive attention mechanism to retain essential features during the distillation process, improving model adaptability in cloud-edge collaborative environments. For the first time, We propose an optimization method based on diffusion model classification loss to realize backdoor attack in data set distillation in edge environment. Experimental results demonstrate that our method improves the accuracy of distilled datasets by 2.8% to 11.7% compared to traditional DNN data distillation methods and by 0.9% to 4.1% compared to existing diffusion model-based data distillation methods. Additionally, our attack methodology achieves backdoor attacks while maintaining an accuracy

关键词： Data models Collaboration Training Synthetic data Diffusion models Computational modeling Cloud computing Accuracy Image edge detection Load modeling Diffusion model data distillation backdoor attacks

来源：评论

学校读者我要写书评

暂无评论

Delay- and Resource-Aware Satellite UPF Service Optimization

引用

IEEE TRANSACTIONS ON MOBILE COMPUTING 2025年第4期24卷 2564-2579页

作者： Wang, Chao Ma, Xiao Xing, Ruolin Li, Sisi Zhou, Ao Wang, Shangguang Beijing Univ Posts & Telecommun State Key Lab Networking & Switching Technol Beijing 100876 Peoples R China

Executing 5G core network functions on satellites has become crucial to enhance satellite network management and service capabilities. The User Plane Function (UPF) is responsible for efficient data traffic forwarding and is envisioned as a key and pioneering core network function that will be deployed on satellites. However, managing and providing services with satellite UPFs face dual challenges. Limited satellite resources constrain the user scale that a satellite UPF can service, resulting in an unguaranteed service delay. Moreover, the extremely rapid mobility of satellites renders it difficult for satellite UPFs to provide seamless services. To address the above challenges, this paper presents the first-of-its-kind service optimization scheme for satellite UPFs in terms of switch control, state migration, and traffic routing. To provide guaranteed service delay, we provide a theoretical analysis based on the M/G/1 queue model, demonstrating the service delay-resource consumption trade-off. A satellite UPF switch control scheme is integrated into the service optimization process, which can decrease satellite UPF service delay while saving satellite resources by adjusting the switch control parameters. To provide seamless services, we propose a satellite UPF-oriented state-aware service migration and traffic routing (UPF service optimization) algorithm. A policy network-based reinforcement learning approach is employed to dynamically perceive the satellite network's state as well as the satellite UPF switch state. Building upon the optimization of service delay through satellite UPF switch control, the processes of state-aware state migration and traffic routing are further employed to reduce delay, ensuring seamless service effectively. Experiments reveal that the proposed algorithm outperforms other benchmark algorithms under different metrics. The service delay is reduced by an average of 23.2% compared with other algorithms.

关键词： Satellites Switches Routing Optimization Delays Control systems Process control Low earth orbit satellites Base stations 5G mobile communication Satellite UPF service optimization delay-resource consumption trade-off policy network

来源：评论

学校读者我要写书评

暂无评论

Part-Level Relationship Learning for Fine-Grained Few-Shot Image Classification

引用

IEEE TRANSACTIONS ON MULTIMEDIA 2025年 27卷 1448-1460页

作者： Wang, Chuanming Fu, Huiyuan Liu, Peiye Ma, Huadong Beijing Univ Posts & Telecommun State Key Lab Networking & Switching Technol Beijing 100876 Peoples R China

Recently, an increasing number of few-shot image classification methods have been proposed, and they aim at seeking a learning paradigm to train a high-performance classification model with limited labeled samples. However, the neglect of part-level relationships causes few-shot methods to struggle to distinguish between closely similar subcategories, which makes it difficult for them to solve the fine-grained image classification problem. To tackle this challenging task, this paper proposes a fine-grained few-shot image classification method that exploits both intra-part and inter-part relationships among different samples. To establish comprehensive relationships, we first extract multiple discriminative descriptors from the input image, representing its different parts. Then, we propose to define the metric spaces by interpolating intra-part relationships, which can help the model adaptively find clear boundaries for these confusing classes. Finally, since the unlabeled image has high similarities to all classes, we project these similarities into a high-dimension space according to the inter-part relationship and interpolate a parameterized classifier to discover the subtle differences among these similar classes. To evaluate our proposed method, we conduct extensive experiments on various fine-grained datasets. Without any pre-train/fine-tuning process, our approach clearly outperforms previous few-shot learning methods, which demonstrates the effectiveness of our approach.

关键词： Measurement Feature extraction Image classification Extraterrestrial measurements Few shot learning Prototypes Neural networks Metalearning Image color analysis Hands Fine-grained image classification few-shot learning metric learning part-level relationship learning

来源：评论

学校读者我要写书评

暂无评论

Growth-adaptive distillation compressed fusion model for network traffic identification based on IoT cloud-edge collaboration

引用

AD HOC NETWORKS 2025年 167卷

作者： Yang, Yang Fan, Chengwen Chen, Shaoyin Gao, Zhipeng Rui, Lanlan Beijing Univ Posts & Telecommun State Key Lab Networking & Switching Technol Beijing 100876 Peoples R China

The development of the Internet of Things (IoT) has led to the rapid growth of the types and number of connected devices and has generated large amounts of complex and diverse traffic data. Traffic identification on edge servers solves the real-time and privacy requirements of IoT management and has attracted much attention, but still faces several problems: (1) traditional machine learning (ML) models rely on artificially constructed features, and the existing deep learning (DL) traffic identification models have reached their performance limit;and (2) insufficient computing resources of edge servers limit the possible improvement in the performance of deep learning models by increasing the number of parameters and structural complexity. To address these issues, we propose a lightweight fusion model. First, the Network-in-Network (NiN) model and Random Forest (RF) model are used on the cloud server to construct a traffic identification fusion model. The excellent representation extraction capability of the NiN compensates for the RF's dependence on manual feature extraction, and its modular structure is suitable for the subsequent model compression operations. Then, the NiN was distilled. We propose Growth-Adaptive Distillation to lightweight the NiN model, which can reduce the operation of manually adjusting the structure of the student model and ensure the efficiency and low power consumption of the fusion model deployment. In addition, both the RF in the cloud and the distilled NiN are deployed on the edge server. Comparisons with multiple algorithms on two network traffic datasets show that the proposed model achieves state-of-the-art performance while ensuring the use of minimal computational resources.

关键词： IoT Traffic identification Network-in-network Random forest Knowledge distillation Fusion model

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：