Title Page
Abstract
요약
Contents
Chapter 1. Introduction 14
1.1. Motivation 15
1.2. Outline of the Thesis and Contributions 17
Chapter 2. Literature Review of Core System Components 20
2.1. Industrial Internet of Things (IIoT) 20
2.2. Internet of Vehicle Edge Computing (IoVEC) Networks 22
2.3. Edge Computing Technology 23
2.4. Federated Learning 24
2.4.1. Federated Averaging (FedAvg) 26
2.4.2. Categorization of Federated Learning 26
2.5. Blockchain: Distributed Ledger Technology 27
2.5.1. Types of Blockchain 29
2.5.2. Smart Contracts 29
2.5.3. Consensus Algorithm 30
2.6. Edge Intelligence System: Concept and Research Trends 31
2.6.1. The Concept of Edge Intelligence System 31
2.6.2. Research Trends on EIS 31
Chapter 3. Trusted Data Management with Decentralized Incentive Mechanism: A Fair Transactions in IoVEC Networks 34
3.1. Introduction 34
3.2. Related Work 37
3.3. Proposed Architecture 38
3.3.1. Entities of Proposed Scheme 39
3.3.2. Design Architecture and Procedures 40
3.4. Performance Evaluation 48
3.4.1. Message Credibility and Rating 50
3.4.2. Block Generation 52
3.4.3. Application Rate 53
3.4.4. Mac/Phy Overhead 54
3.4.5. Packet Delivery Ration (PDR) 55
3.4.6. Decentralized Incentive Mechanism 56
3.5. Security Analysis and Discussion 59
(ⅰ) Decentralized Approach without an Intermediary 59
(ⅱ) Non-Repudiation and Data Consistency 59
(ⅲ) Tamper-Proofing and Data Unforgeability 60
(ⅳ) Data Availability 60
(ⅴ) Data Privacy and Credibility 60
3.6. Discussion 61
3.7. Conclusion 63
Chapter 4. An Improved Privacy-preserving Edge Intelligence using Differential Privacy 64
4.1. Introduction 64
4.2. Fundamentals of Differential Privacy 66
4.3. Towards Secure Edge Intelligence 67
4.3.1. Design Architecture Overview: A Practical Example of IoVEC 68
4.3.2. Workflow of Proposed Architecture 69
4.4. Analysis of Simulation Results 73
4.4.1. LDP-based FL simulation 74
4.4.2. Blockchain-based FL simulation 77
4.5. Conclusion 79
Chapter 5. Enhancing Personalized Federated Learning with Decentralized Edge Clustering for Heterogeneous Data 81
5.1. Introduction 81
5.2. Background 83
5.2.1. The Effect of Heterogeneous Data in Federated Learning 83
5.2.2. The Concept of Personalized Federated Learning (PFL) 83
5.3. Works on Personalized FL 84
5.4. Decentralized Edge Cluster for Personalized FL 86
5.4.1. System Initialization 88
5.4.2. Collaborative Edge Cluster Establishment and Personalized Local Training Models 89
5.4.3. Personalized Global Model Aggregation and Incentive Mechanism 91
5.5. Numerical Results and Discussion 92
5.5.1. Implementation 92
5.5.2. Discussion 94
5.6. Conclusion 96
Chapter 6. Conclusions and Recommendations 97
6.1. Conclusions 97
6.2. Recommendations 98
Bibliography 100
List of Publications 114
Glossary 119
TABLE 2.1. Comparison of Consensus Algorithms 30
TABLE 3.1. Summary of Notations 41
TABLE 3.2. Simulation Parameters 50
TABLE 3.3. Result of message credibility assessment of MVP[이미지참조] 51
TABLE 3.4. Performance of the total gas usage and Ether distribution for the packet delivery ratio provider. 58
TABLE 3.5. The key parameter comparison of our proposed system with other solutions. 61
TABLE 4.1. Impact of privacy levels (є) on system accuracy. 76
TABLE 4.2. Comparison of the proposed system with related works. 80
TABLE 5.1. Summary of the state-of-the-art methods for the PFL research. 86
TABLE 5.2. Summary of BPFL advantages compared to current PFL techniques. 95
FIGURE 1.1. Diagram depicts the relationship between one chapter to the others 18
FIGURE 2.1. Illustration of IIoT applications in various industries. 21
FIGURE 2.2. Illustration of IoVEC architecture. 22
FIGURE 2.3. Illustration of FedAvg procedures. 25
FIGURE 2.4. Illustration of blockchain structure. 27
FIGURE 3.1. Overview of design architecture. 39
FIGURE 3.2. Consensus process in validation block smart contracts (VBSC). 45
FIGURE 3.3. Block structure of transaction. 47
FIGURE 3.4. Illustration of incentive mechanism. 48
FIGURE 3.5. Simulation scenario map. 49
FIGURE 3.6. Trust value rating aggregation based on message credibility assess- ment by VAn.[이미지참조] 51
FIGURE 3.7. Message credibility rating versus distance of VP to occurred event.[이미지참조] 52
FIGURE 3.8. Effect of batch size on throughput. 53
FIGURE 3.9. Application packets receiving rate. 54
FIGURE 3.10. MAC/PHY overhead. 55
FIGURE 3.11. Packet delivery ratio over RSUs. 56
FIGURE 3.12. The information on gas usage by RSU in distributing Ether for VPKn[이미지참조] 58
FIGURE 4.1. Overview of the joint framework for decentralized EIS. 69
FIGURE 4.2. Workflow of the proposed architecture. 70
FIGURE 4.3. Accuracy comparison of various privacy budget. 75
FIGURE 4.4. Loss function value for various numbers of vehicles. 78
FIGURE 4.5. Smart contracts' initial migration and deployment. 78
FIGURE 4.6. Distribution of edge servers' contributions. 79
FIGURE 4.7. Performance accuracy comparison. 79
FIGURE 5.1. Decentralized edge cluster for personalized federated learning. 87
FIGURE 5.2. Training loss of 50 epochs for MLP and CNN models. 92
FIGURE 5.3. Accuracy performance results on IID and Non-IID setting. 93
FIGURE 5.4. Distribution of edge cluster contributions towards generating the global model PFL based on Ethereum platform. 94
FIGURE 5.5. Performance accuracy comparison. 95