检索结果-内蒙古大学图书馆

WFB: watermarking-based copyright protection framework for federated learning model via blockchain

SCIENTIFIC REPORTS 2024年第1期14卷 1-15页

作者： Shao, Sujie Wang, Yue Yang, Chao Liu, Yan Chen, Xingyu Qi, Feng Beijing Univ Posts & Telecommun State Key Lab Networking & Switching Technol Beijing 100876 Peoples R China State Grid Liaoning Elect Power Co Ltd Shenyang 110006 Peoples R China

Federated learning (FL) enables users to train the global model cooperatively without exposing their private data across the engaged parties, which is widely used in privacy-sensitive business. However, during the life cycle of FL models, both adversaries' attacks and ownership generalization threaten the FL models' copyright and affect the models' reliability. To address these problems, existing model watermarking techniques can be used to verify FL model's ownership. However, due to the lack of credible binding from "model extracted watermarks" to "ownership verification", it is difficult to form a closed-loop watermarking framework for copyright protection. Therefore, starting from the shortcomings of the current watermark verification scheme, this article proposed WFB, a blockchain-empowered watermarking framework for ownership verification of federated models. Firstly, we propose a improved watermark generation algorithm to solve the credibility issue of watermarks. Secondly, we propose a watermark embedding method in federated learning, while blockchain technology is used to ensure the credible storage of watermark information throughout the process. Thirdly, the credibility of ownership verification is improved because of the watermark authenticity. Experimental results demonstrate the fidelity, effectiveness and robustness of WFB, with other superiorities such as improving process security and traceability.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Device-collaborative Computation Offloading in Mobile Edge Networks Based on Propagation Dynamics Theory 2

Device-collaborative Computation Offloading in Mobile Edge N...

引用

2nd International Joint Conference on Information and Communication Engineering, JCICE 2023

作者： Ren, Shuidi Xu, Xiaodong Han, Shujun Beijing University of Posts and Telecommunications State Key Laboratory of Networking and Switching Technology Beijing China

ISBN: (纸本)9798350325768

In the future Internet of Intelligent Things, when multiple devices have too much computation task at one moment, device-collaborative computation offloading helps them to reduce the task pressure while it may cause more users to suffer from task overload due to computation task delivery. Finding the critical conditions that lead to the all-user task overload, it helps us deploy the communication network to guarantee its stability. Therefore, in this paper we study the process of computation task delivery among the mobile edge networks (MENs) and its impact. We utilize the spreading threshold model in propagation dynamics theory to characterize the device-collaborative computation offloading. The edge-based compartmental theory is applied to describe the proposed model, establish the propagation dynamics equations, and derive the saddle points. Through analysis and simulation, we find that the network status that all users suffer from data overload can be avoided by controlling the proportion of users who choose device-collaborative computation offloading and the network heterogeneity. © 2023 IEEE.

关键词： Computation offloading

来源：评论

学校读者我要写书评

暂无评论

Minimizing Adversarial Training Samples for Robust Image Classifiers: Analysis and Adversarial Example Generator Design

引用

IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY 2024年 19卷 9613-9628页

作者： Wang, Yulong Sun, Tong Yuan, Xin Li, Shenghong Ni, Wei Beijing Univ Posts & Telecommun State Key Lab Networking & Switching Technol Sch Comp Sci Natl Pilot Software Engn Sch Beijing 100000 Peoples R China CSIRO Sydney NSW 2122 Australia

Training deep neural networks (DNNs) with altered data, known as adversarial training, is essential for improving their robustness. A significant challenge emerges as the robustness strengthened during training often diminishes during inference, resulting in drops in robust pronounced accuracy. Contemporary strategies either necessitate excessively large training data or risk compromising the natural accuracy of non-adversarial images. Our analysis identifies that the inherent vulnerability of DNNs to adversarial attacks stems from certain input space segments that are inadequately populated by training data, leading to decision-making voids with incorrect predictions. The minimum number of training samples required for successful adversarial training can be attained by maximizing the representativeness of the samples. In light of this, we put forth an advanced training data augmentation method anchored on a Generative Adversarial Network. The generated samples are evaluated by the image classifier during training and selected based on their confidence scores. Evaluations on public datasets, such as Tiny-ImageNet, MS COCO, and CIFAR-100, using various deep neural networks (DNNs), including Vision Transformer, MobileNet, and WideResNet, under recent attacks like DifAttack, SQBA, and AutoAttack, confirm that our method significantly enhances the adversarial robustness of DNN image classifiers. Our method outperforms state-of-the-art adversarial training methods by 35.66% on Tiny-ImageNet, 13.53% on MS COCO, and 13.06% on CIFAR-100.

关键词： Deep neural network adversarial training generative adversarial network.

来源：评论

学校读者我要写书评

暂无评论

Cross Attention Scoring Function for Multimedia Domain Knowledge Graph Completion

Cross Attention Scoring Function for Multimedia Domain Knowl...

引用

2023 IEEE International Symposium on Broadband Multimedia Systems and Broadcasting, BMSB 2023

作者： Fang, Honglin Yu, Peng Wang, Ying Li, Wenjing Beijing University of Posts and Telecommunications State Key Laboratory of Networking and Switching Technology Beijing China

ISBN: (纸本)9798350321524

Multimedia-based applications and services are strongly expanded by the continuous development of communication and network technologies. Recently, with the high efficiency of knowledge graph(KG) in data modeling, various KGs in multimedia domains have been constructed to manage and utilize multimedia services. In multimedia KGs, the integrity of the link structure can significantly impact the performance of downstream applications. Numerous knowledge graph completion(KGC) models are proposed to predict lost relations between KG entities. However, these general models target at capturing different relation patterns and lack the attention of the relationship between entities and relations. In this paper, a novel KGC model named TransX is proposed to learn this complex inner relationship, which can robustly reflect the link structure of the multimedia domain KGs. TransX firstly projects entities and relations to multiple property spaces, and then a new cross attention block is designed to represent different patterns between entity and relation properties. The experiment results demonstrate that TransX has fewer parameters and better performance at the cost of introducing some sufferable computation in the cross attention block. © 2023 IEEE.

关键词： Knowledge graph

来源：评论

学校读者我要写书评

暂无评论

A Multiple QoS Metrics-Aware Virtual Network Embedding Algorithm

引用

COMPUTER JOURNAL 2023年第3期67卷 1171-1186页

作者： Lu, Meilian Li, Meng Beijing Univ Posts & Telecommun Sch Comp Sci Natl Pilot Software Engn Sch State Key Lab Networking & Switching Technol Beijing 100876 Peoples R China

As a key issue of network virtualisation, virtual network embedding (VNE) aims to embed multiple virtual network requests (VNRs) from different applications onto the substrate network effectively. In real networks, about 90% of traffic is generated by different quality of service (QoS) sensitive applications. However, most existing VNE algorithms do not account for the difference. Although several VNE algorithms considered the delay metric of applications, they usually provide strict delay guarantees for all VNRs, leading to a low VNR acceptance ratio. In this paper, we focus on the VNE problem involving multiple QoS metrics and propose a multiple QoS metrics-aware VNE algorithm based on reinforcement learning (RLQ-VNE). We first classify VNRs according to their different requirements for multiple QoS metrics including delay, jitter and packet loss rate, and then introduce reinforcement learning to implement differentiated VNE. Specifically, RLQ-VNE provides strict QoS guarantees for the VNRs with high-level QoS requirements and provides lower QoS guarantees for the VNRs with low-level QoS requirements, thus balancing the QoS guarantee and request acceptance ratio. Simulation results from multiple experimental scenarios show that RLQ-VNE improves the request acceptance ratio and network resource utilisation by sacrificing less QoS.

关键词： virtual network embedding quality of service differentiated service reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

On Optimal Code Hopping Pattern Design for Complementary Coded CHMA Systems

引用

IEEE TRANSACTIONS ON VEHICULAR technology 2024年第12期73卷 18988-19004页

作者： Liu, Xiqing Yang, Yiran Lai, Guan-Lun Chen, Hsiao-Hwa Beijing Univ Posts & Telecommun State Key Lab Networking & Switching Technol Beijing 100876 Peoples R China Natl Cheng Kung Univ Dept Engn Sci Tainan 70101 Taiwan

Code hopping multiple access (CHMA) is an emerging technology to implement code domain non-orthogonal multiple access (NOMA). In CHMA, multiple access interference and multipath interference can be mitigated with the help of complementary codes, namely complementary coded CHMA (CC-CHMA). However, traditional CC-CHMA works on random code hopping (CH) patterns, which leads to an irreducibly high bit error floor. To address this issue, this work proposes an optimal CH pattern design to minimize the bit error floor. In addition, code collision probability and bit error rate (BER) with binary phase shift keying modulation are derived explicitly. The results verify that CC-CHMA with the proposed optimal CH pattern exhibits a much better BER performance than traditional CC-CHMA, which at the same time also offers a higher overload rate than traditional NOMA in uplink multipath channels, paving a way for its possible applications in beyond 5G mobile communications.

关键词： Codes Symbols Downlink Multiaccess communication NOMA Interference Generators Code hopping multiple access non-orthogonal multiple access complementary code code collision code hopping pattern

来源：评论

学校读者我要写书评

暂无评论

Multi-Relational Knowledge Enhanced Contrastive Learning for Event Extraction 9

Multi-Relational Knowledge Enhanced Contrastive Learning for...

引用

9th International Conference on Computer and Communications, ICCC 2023

作者： Liu, Yixuan Shuang, Kai Beijing University of Posts and Telecommunications State Key Laboratory of Networking and Switching Technology Beijing China

ISBN: (纸本)9798350317251

Event extraction (EE) is a fundamental task under the field of natural language processing. Most of the existing EE methods perform supervised fine-tuning, training downstream with the help of the ability to pre-train language models (PLMs), limited labeled data is difficult to satisfy the training needs of PLMs and faces problems such as limited generalization. Contrastive learning, as a self-supervised training paradigm, improves the representational power of PLMs by acquiring a wider range of knowledge from large-scale unlabeled corpora than is implied by the data labels. However, existing contrastive learning methods mainly focus on the sentence-level, while the development at the token-level represented by EE tasks is still in its early stages, resulting in noisy representations of tokens and a lack of detailed use of event knowledge, which makes the performance bottleneck of the PLMs after contrastive learning when fine-tuning downstream EE tasks. To this end, we propose MRKE-CLEE, a multi-relational knowledge enhanced contrastive learning for event extraction. MRKE-CLEE first accomplishes the coreference resolution and AMR parsing as supported tasks to obtain multiple event-relationship knowledge pseudo-labels from a large-scale unlabeled corpus through the auxiliary model, then uses in-sentence knowledge guided contrast to filter noisy contrastive pairs to make the resulting token representations more distinct, and using event type knowledge guided contrast to aggregate token representations with the same event type. With the combination of multi-relational event knowledge, using contrastive learning to improve the performance of event detection (ED) and event argument extraction (EAE). Experiments demonstrate that our MRKE-CLEE achieves a new state-of-the-art performance on the ACE2005 and MAVEN datasets and introduces new attempts to use token-level contrastive learning in EE tasks. © 2023 IEEE.

关键词： Extraction

来源：评论

学校读者我要写书评

暂无评论

BlindSpotEliminator: Collaborative Point Cloud Perception in Cellular-V2X Networks

BlindSpotEliminator: Collaborative Point Cloud Perception in...

引用

2023 IEEE International Conference on Systems, Man, and Cybernetics, SMC 2023

作者： Chen, Ziyue Luo, Guiyang Shao, Congzhang Yuan, Quan Li, Jinglin Beijing University of Posts and Telecommunications State Key Laboratory of Networking and Switching Technology Beijing China

ISBN: (纸本)9798350337020

Multi-agent collaborative perception depends on sharing sensory information to improve perception accuracy and robustness, as well as to extend coverage. However, most collaborative perception methods ignore the limitations of communication networks, such as limited bandwidth and the possibility of wireless conflicts. To fill this gap, this paper proposes BlindSpotEliminator, a conflict-free scheduler over the cellular-V2X networks for supporting practical collaborative point cloud perception to eliminate blind spots. BlindSpotEliminator first identifies the blind spots for each vehicle, then lists the corresponding conflict relationships based on the distribution of the blind spots and communication conflicts, and finally designs an optimized point cloud data transmission strategy to eliminate the blind spots of each vehicle. Extensive experiments show that compared with greedy algorithm and random methods, BlindSpotEliminator achieves better efficiency, i.e., transmitting 20% more point cloud data. © 2023 IEEE.

关键词： Cellular V2X collaborative perception optimization strategy point cloud data transmission

来源：评论

学校读者我要写书评

暂无评论

A Federated Learning Approach for Net Load Forecasting in Microgrids. 24

A Federated Learning Approach for Net Load Forecasting in Mi...

引用

24th Asia-Pacific Network Operations and Management Symposium, APNOMS 2023

作者： Qin, Zihan Shao, Sujie Guo, Shaoyong Qiu, Xuesong State Key Laboratory of Networking and Switching Technology Beijing University of Posts and Telecommunications Beijing China

ISBN: (纸本)9788995004395

Microgrid is a small-scale power grid composed of various distributed energy resources, and net load forecasting (NLF) plays significant role in its energy generation planning. Due to the dispersed distribution of nodes within the microgrid and disparate user data, completing the net load forecasting task becomes challenging. In order to efficiently forecast the net load while ensuring data security, this paper proposes an approach for NLF based on federated learning (FL). Firstly, a FL framework for NLF in microgrids is constructed to complete the NLF task with long short-term memory neural network (LSTM) model and node selection (NS) in aggregation. Secondly, considering factors such as the data precision and quantity, an accuracy optimization problem model is proposed. Then, the problem is formulated as a Markov decision process (MDP), and a NS algorithm based on Asynchronous Advantage Actor-Critic (A3C) is designed for selecting a node set to optimize model aggregation. The simulation demonstrate that the proposed NS method based on A3C improves the performance of FL, resulting in better convergence of LSTM network and ensuring the scalability of the cluster in microgrid. Copyright 2023 KICS.

关键词： Long short-term memory

来源：评论

学校读者我要写书评

暂无评论

Boosting DNN Cold Inference on Edge Devices 23

Boosting DNN Cold Inference on Edge Devices

引用

21st Annual International Conference on Mobile Systems, Applications and Services, MobiSys 2023

作者： Yi, Rongjie Cao, Ting Zhou, Ao Ma, Xiao Wang, Shangguang Xu, Mengwei State Key Laboratory of Networking and Switching Technology Beijing China Microsoft Research Beijing China

ISBN: (纸本)9798400701108

DNNs are ubiquitous on edge devices nowadays. With its increasing importance and use cases, it's not likely to pack all DNNs into device memory and expect that each inference has been warmed up. Therefore, cold inference, the process to read, initialize, and execute a DNN model, is becoming commonplace and its performance is urgently demanded to be optimized. To this end, we present NNV12, the first on-device inference engine optimizing cold inference. NNV12 is built atop three novel optimization knobs: selecting a proper kernel (i.e., operator implementation) for each DNN operator, bypassing the weights transformation process by caching the post-transformed weights on disk, and pipelined execution of many kernels on asymmetric processors. To tackle with the huge search space, NNV12 employs a heuristic-based scheme to obtain a near-optimal kernel scheduling plan. We fully implement a prototype of NNV12 and evaluate its performance across extensive experiments. It shows that NNV12 achieves up to 15.2× speedup compared to the state-of-the-art DNN engines on edge CPUs and 401.5× speedup on edge GPUs, respectively. © 2023 Owner/Author(s).

关键词： Engines

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：