检索结果-内蒙古大学图书馆

3rd International Conference on Algorithms, Data Mining, and Information Technology, ADMIT 2024

作者： Chen, Zhouguo Chen, Hongda Liu, Shuai Liu, Yongdong Schoolof Computer Science and Engineering Southeast University Jiangsu Nanjing China The 30th Research Institute of China Electronics Technology Group Corporation Sichuan Chengdu China School of Software Engineering Xian Jiaotong University Shaanxi Xian China China Industrial Control Systems Cyber Emergency Response Team Beijing China

ISBN: (纸本)9798400718120

Federated Learning (FL) has gained considerable attention for collaborative training in big data analysis, particularly in terms of privacy and communication constraints. Despite its promising advantages, FL faces the challenge of dealing with imbalanced label distribution on edge devices, which complicates the learning task. In literature, numerous studies have attempted to tackle this issue by introducing additional hyperparameters, which can complicate and prolong the process of model training and tuning. This complexity may potentially lead to overfitting and limited performance on new datasets. In this paper, we present a new Federated Learning framework called FedVC, which introduces virtual clients to minimize hyperparameter reliance. FedVC aims to bridge the gap between balanced and imbalanced label distribution. Compared with traditional federated learning frameworks, the proposed FedVC model designs a dynamically balanced dataset by utilizing virtual clients, which is proposed for the first time in this work. These virtual clients determine which part of the data should be used in one single training round and manage backpropagation timing for local real clients. The balanced dataset promotes greater stability in local models, thereby enhancing the overall stability of the training process. The experimental results show that FedVC outperforms classical FedAvg and the compared baseline methods for the data with imbalanced label distribution. Especially, FedVC exhibits more stable and accurate results than alternative approaches in highly heterogeneous settings. The source code is available at https://***/aikedaerC/FedVC. © 2024 Copyright held by the owner/author(s).

关键词： Federated learning

来源：评论

学校读者我要写书评

暂无评论

A Method for Extracting Small Water Areas Based on an Improved FPN Network

A Method for Extracting Small Water Areas Based on an Improv...

引用

International Conference on Ubiquitous Information Management and Communication (IMCOM)

作者： Yufeng Yu Meixiang Wang Qun Zhao Yong Wang Qingqing Chen Department of Computer and Software Hohai University Nanjing Jiangsu China School of Computer Engineering Nanjing Institute of Technology Nanjing Jiangsu China Bank of Nanjing Co. Jiangsu Nanjing China Aerosun Corporation Information Center Nanjing Jiangsu China

ISBN: (数字)9798331507817

ISBN: (纸本)9798331507824

Small water areas such as river branches, lakes, and ponds are vital components of aquatic ecosystems closely linked to human activities. Extracting small water areas from remote sensing images holds significant importance in water resource planning, investigation, sustainable water resource management, and environmental protection. However, the current small water area extraction methods based on remote sensing often encounter the problems of small target, few features, difficult localization and easy loss of edge features. Therefore, this paper proposes a small water area extraction model based on IFPN_BEM network. The model leverages a Transformer Encoder module with deformable attention mechanisms to globally model deep features of the FPN network and optimizes small water area edges using the BEM module, which further improves the extraction accuracy of small water areas. Experimental results on the GID dataset demonstrate that this model provides more complete and accurate extraction of small water areas, achieving precision, F1 score, and IoU accuracy metrics of 81.3%, 84.4%, and 73.4% respectively. These metrics show improvements of 9.5%, 8.2%, and 8.5% over the original FPN network, outperforming UNet, DeepLabV3+, and the improved FPN + PointRend networks across all metrics.

关键词： Measurement Deformable models Accuracy Attention mechanisms Image edge detection Feature extraction Transformers Data mining Water resources Remote sensing

来源：评论

学校读者我要写书评

暂无评论

Machine Learning and Deep Learning Methods on Breast Cancer Metastases Detection 4

Machine Learning and Deep Learning Methods on Breast Cancer ...

引用

4th International Conference on Computing and Data Science, CONF-CDS 2022

作者： Zheng, Wentao Liu, Kan Che, Fuqi Zheng, Yichen School of Computer Science and Engineering Central South University Hunan Changsha410083 China School of Optics and Photonics Beijing Institute of Technology Beijing100089 China School of Software Dalian University of Technology Liaoning Dalian116620 China Dulwich International High School Jiangsu Suzhou215021 China

There is a lot of room for research in how to detect breast cancer metastases in whole slide images more quickly and precisely. This work contrasts several machine learning and deep learning models on the breast cancer metastases detection problem and experiments the impacts of feature extraction, dimension reduction and the sizes of the datasets on these models. This work also establishes a breast cancer metastases detection pipeline based on the most accurate model in the experiments. Among the models, Dense Model using features trained from DenseNet201 with weights trained by ImageNet is of the best performance. This work develops a computer aided method to examine the cancerous regions in the whole slide images of breast cancer more quickly and precisely than an experienced doctor does. © Published under licence by IOP Publishing Ltd.

关键词： Feature extraction

来源：评论

学校读者我要写书评

暂无评论

Martingale Theory Based Definition and Analysis of Energy Self-Sustainability in Batteryless Internet of Things

IEEE Transactions on Green Communications and Networking

引用

IEEE Transactions on Green Communications and Networking 2024年

作者： Fan, Xinyu Hu, Jie Yang, Kun University of Electronic Science and Technology of China School of Information and Communication Engineering Chengdu611731 China Nanjing University School of Intelligent Software and Engineering Jiangsu Suzhou China University of Essex School of Computer Science and Electronic Engineering United Kingdom

Radio-frequency (RF) based Wireless Energy Transfer (WET) enables devices to avoid wired power-supply and battery replacements. The ubiquitous availability of RF energy elevates 'energy self-sustainability' as a pivotal goal for wireless sensor networks. However, for RF energy harvesting networks, the term 'energy self-sustainability"lacks a precise mathematical characterization. This paper presents a robust mathematical definition for energy self-sustainability within integrated data and energy networks. By utilizing Martingale theory, we develop a mathematical framework that determines the energy self-sustainability, by acquiring the stochastic properties of energy harvesting and consumption processes. The fundamental paradigm of utilizing this framework to derive energy self-sustainability is demonstrated. In-depth explorations have been conducted on the diverse stochastic characteristics of energy harvesting and consuming processes. Additionally, this study delves into the anticipated uninterrupted operating duration and associated energy expectation. Monte-Carlo simulations confirm the precision of our theoretical evaluations. By analyzing the correlation between harvested and consumed energy in the context of varied energy self-sustainability requirements, this paper provides design guidance for energy transmitters and batteryless wireless devices. © 2017 IEEE.

关键词： Stochastic systems

来源：评论

学校读者我要写书评

暂无评论

Fast Mesh Reconstruction from Single View Based on GCN and Topology Modification

引用

computer Systems Science & engineering 2023年第5期45卷 1695-1709页

作者： Xiaorui Zhang Feng Xu Wei Sun Yan Jiang Yi Cao Wuxi Research Institute Nanjing University of Information Science&TechnologyWuxi214100China Engineering Research Center of Digital Forensics Ministry of EducationJiangsu Engineering Center of Network MonitoringSchool of Computer and SoftwareNanjing University of Information Science&TechnologyNanjing210044China Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology(CICAEET) Nanjing University of Information Science&TechnologyNanjing210044China School of Automation Nanjing University of Information Science&TechnologyNanjing210044China Department of Electrical and Computer Engineering University of WindsorWindsorONN9B 3P4Canada

3D reconstruction based on single view aims to reconstruct the entire 3D shape of an object from one *** existing methods reconstruct the mesh surface of complex objects,the surface details are difficult to predict and the reconstruction visual effect is poor because the mesh representation is not easily integrated into the deep learning framework;the 3D topology is easily limited by predefined templates and inflexible,and unnecessary mesh self-intersections and connections will be generated when reconstructing complex topology,thus destroying the surface details;the training of the reconstruction network is limited by the large amount of information attached to the mesh vertices,and the training time of the reconstructed network is too *** this paper,we propose a method for fast mesh reconstruction from single view based on Graph Convolutional Network(GCN)and topology *** use GCN to ensure the generation of high-quality mesh surfaces and use topology modification to improve the flexibility of the ***,a feature fusion method is proposed to make full use of the features of each stage of the image *** use 3D open dataset ShapeNet to train our network and add a new weight parameter to speed up the training *** experiments demonstrate that our method can not only reconstruct object meshes on complex topological surfaces,but also has better qualitative and quantitative results.

关键词： 3D surface reconstruction deep learning GCN topology modification end-to-end framework

来源：评论

学校读者我要写书评

暂无评论

Private Data Aggregation Enabling Verifiable Multisubset Dynamic Billing in Smart Grids

Private Data Aggregation Enabling Verifiable Multisubset Dyn...

引用

IEEE International Conference on Trust, Security and Privacy in Computing and Communications (TrustCom)

作者： Qian Yang Chen Wang Jian Shen Yi Li Dengzhi Liu School of Computer Science and Technology Zhejiang Sci-Tech University HangZhou China School of Software and Electrical Engineering Swinburne University of Technology Hawthorn VIC Australia School of Computer Engineering Jiangsu Ocean University Lianyungang China

ISBN: (数字)9798331506209

ISBN: (纸本)9798331506216

Efficient power management in smart grids relies on regularly obtaining the electricity usage of each user. However, the aggregation of the electricity usage data may expose user privacy. At present, most existing solutions aggregate the electricity data of the entire user set, which cannot meet the fine-grained requirements of the control center. Therefore, this paper proposes a verifiable privacy-preserving multisubset dynamic billing data aggregation scheme. Firstly, we divide the electricity data into k consecutive subsets and set dynamic pricing rules, such as pricing within a certain range as q 1 and exceeding it as q 2 . Then, the smart meter encrypts the data using the Paillier cryptographic system with the corresponding subset of parameters and uploads them to the aggregation equipment. After the aggregation equipment completes the users’ data aggregation, the ciphertext C is sent to the control center. This process enables the control center to check data integrity and obtain the total number of people, electricity consumption, and costs for different ranges of electricity in a period of time at once, without the need to obtain data for individual users. Analysis and experiments show that this scheme can resist attacks from powerful adversaries on the transmission channel and has practicality and effectiveness.

关键词： Data privacy Costs Statistical analysis Electricity Pricing Resists Data aggregation Smart meters Smart grids Cryptography

来源：评论

学校读者我要写书评

暂无评论

Ensemble Deep Learning Based Single Finger-Vein Recognition 6th

Ensemble Deep Learning Based Single Finger-Vein Recognition

引用

6th International Conference on Cognitive Systems and Signal Processing, ICCSIP 2021

作者： Liu, Chongwen Qin, Huafeng Yang, Gongping Shen, Zhengwen Wang, Jun School of Computer Science and Information Engineering Chongqing Technology and Business University Chongqing400067 China School of Software Engineering Shandong University Jinan250101 China China University of Mining and Technology Jiangsu Xuzhou221116 China

ISBN: (纸本)9789811692468

Finger-vein biometrics has been extensively investigated for personal verification. Single sample per person (SSPP) finger-vein recognition is one of the open issues in finger-vein recognition. In this paper, a deep ensemble learning method is proposed for SSPP finger-vein recognition. To the best of our knowledge, this is the first work proposed for finger-vein identification with single finger-vein sample per person. First, we generate different feature maps from an input image, based on which multiple independent deep learning based classifiers are proposed for finger-vein identification. Second, a shared learning scheme is investigated among classifiers to improve their feature representation captivity. Third, an approach is proposed to adjust learning speed of weak classifiers, so that all classifiers achieve best performance at the same time. Finally, a deep ensemble learning method are proposed by combing all weak classifiers. The experimental results on two public finger-vein databases show that our approach outperforms its all classifiers and existing approaches, and achieves the state-of-the-art recognition results. © 2022, Springer Nature Singapore Pte Ltd.

关键词： Learning systems

来源：评论

学校读者我要写书评

暂无评论

引用

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： Xiaomeng Xin Heping Song Jianping Gou School of Computer Science and Communication Engineering Jiangsu University Zhenjiang PR China College of Computer and Information Science College of Software Southwest University Chongqing PR China

The previous relation-based knowledge distillation methods tend to construct global similarity relationship matrix in a mini-batch while ignoring the knowledge of neighbourhood relationship. In this paper, we propose a new similarity-based relational knowledge distillation method that transfers neighbourhood relationship knowledge by selecting K-nearest neighbours for each sample. Our method consists of two components: Neighbourhood Feature Relationship Distillation and Neighbourhood Logits Relationship Distillation. We perform extensive experiments on CIFAR100 and Tiny ImageNet classification datasets and show that our method outperforms the state-of-the-art knowledge distillation methods. Our code is available at: https://***/xinxiaoxiaomeng/***.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Optimizing Federated Learning with Local Differential Privacy: A Game-Theoretic Approach for Privacy-Utility Tradeoff

SSRN

引用

SSRN 2024年

作者： Zeng, Qingkui Yin, Chunyong School of Electronics and Information Engineering Nanjing University of Information Science and Technology Jiangsu Nanjing210044 China School of Computer and Software Nanjing University of Information Science and Technology Jiangsu Nanjing210044 China

The Industrial Internet has brought about privacy concerns alongside productivity improvements. While the emergence of federated learning mitigates the risk of end-device data leakage, it still needs integration with differential privacy to enhance privacy. However, achieving an optimal balance between data utility and privacy in federated learning poses a significant challenge. This paper presents a game-theoretic framework aimed at achieving an optimal trade-off between privacy and utility in federated learning, ensuring users personalized privacy settings and global model performance. Our proposed game model facilitates the allocation of payoffs between users and the server, enabling clients to select privacy budgets based on preferences and payoffs, while the server optimizes privacy and data utility. Through Nash equilibrium, our framework effectively realizes trade-offs between privacy and utility. Experimental validation on three datasets demonstrates that our framework outperforms uniform privacy budget local differential privacy budgeting mechanisms, particularly in terms of the privacy-utility trade-off. © 2024, The Authors. All rights reserved.

关键词： Budget control

来源：评论

学校读者我要写书评

暂无评论

Corporate Fraud Detection in Rich-yet-Noisy Financial Graph

arXiv

引用

arXiv 2025年

作者： Wang, Shiqi Zhang, Zhibo Fang, Libing Nguyen, Cam-Tu Li, Wenzhon School of Artificial Intelligence Nanjing University Jiangsu Nanjing China School of Computer Science and Technology Nanjing University Jiangsu Nanjing China School of Management and Engineering Nanjing University Jiangsu Nanjing China State Key Laboratory for Novel Software Technology Nanjing University Jiangsu Nanjing China

Corporate fraud detection aims to automatically recognize companies that conduct wrongful activities such as fraudulent financial statements or illegal insider trading. Previous learning-based methods fail to effectively integrate rich interactions in the company network. To close this gap, we collect 18-year financial records in China to form three graph datasets with fraud labels. We analyze the characteristics of the financial graphs, highlighting two pronounced issues: (1) information overload: the dominance of (noisy) non-company nodes over company nodes hinders the message-passing process in Graph Convolution Networks (GCN);and (2) hidden fraud: there exists a large percentage of possible undetected violations in the collected data. The hidden fraud problem will introduce noisy labels in the training dataset and compromise fraud detection results. To handle such challenges, we propose a novel graph-based method, namely, Knowledge-enhanced GCN with Robust Two-stage Learning (KeGCNR), which leverages Knowledge Graph Embeddings to mitigate the information overload and effectively learns rich representations. The proposed model adopts a two-stage learning method to enhance robustness against hidden frauds. Extensive experimental results not only confirm the importance of interactions but also show the superiority of KeGCNR over a number of strong baselines in terms of fraud detection effectiveness and robustness. Copyright © 2025, The Authors. All rights reserved.

关键词： Graph embeddings

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：