机器学习中可验证计算的隐私保护技术研究

发布时间：2021-10-25 03:26

　　互联网和物联网的快速发展开启了信息时代的新纪元,数据呈现数量庞大、类型繁多、增速快、价值密度低和真实性等特性。机器学习作为实现人工智能的途径,重点研究如何从海量数据中获取隐藏的、有效的、可理解的知识,建立数据驱动型的推理与决策模型,实现“取之于数据,用之于数据”的目标。然而,传统的机器学习算法通常包含计算密集型的学习过程,对于资源受限的终端用户来说存在应用局限性。此外,训练数据量的匮乏直接导致机器学习模型过拟合或精度低。因此,基于云计算的机器学习技术应运而生,并得到了学术界、产业界和政府的广泛关注。云计算是分布式计算、效用计算、并行计算和虚拟化等多种技术的融合演进和跃升,用户能够以按需付费的方式享受云平台上无尽的存储和计算资源。因而,用户在云服务器的协助下进行模型训练与优化,不仅极大地降低了用户端的计算开销和维护成本,而且可以实现分布式数据集的有效利用。然而,由于用户数据中通常包含敏感信息且云服务器不完全可信,因此,基于云计算的机器学习技术不可避免地面临一些安全问题。首先,数据外包使得用户失去了对其物理管控,如何保证训练过程中训练集数据隐私性和计算结果的可验证性,是面临的安全挑战之一。... 

【文章来源】：西安电子科技大学陕西省 211工程院校 教育部直属院校

【文章页数】：131 页

【学位级别】：博士

【文章目录】：
摘要
ABSTRACT
List of Abbreviations
Chapter 1 Introduction
    1.1 B ackground
    1.2 Related Work
    1.3 Our Contributions
    1.4 Organization
Chapter 2 Preliminaries
    2.1 Bilinear Pairings
    2.2 Paillier Encryption
    2.3 Publicly Verifiable Computation
    2.4 Summary
Chapter 3 Privacy-preserving and Publicly Verifiable Matrix Multiplication SchemeDeployed in Machine Learning
    3.1 Overview
    3.2 Problem Statement
        3.2.1 System Model
        3.2.2 Treat Model
    3.3 The Proposed Scheme
        3.3.1 Basic Components
        3.3.2 Main Idea
        3.3.3 The Concrete Construction
        3.3.4 Correctness
    3.4 Security Analysis
    3.5 Performance Analysis
        3.5.1 Efficiency Analysis
        3.5.2 Experimental Evaluation
    3.6 Summary
Chapter 4 Privacy-preserving and Verifiable SLP Training and Prediction Scheme
    4.1 Overview
    4.2 Problem Statement
        4.2.1 System Model
        4.2.2 Definition of Privacy-preserving Outsourcing Matrix Multiplication
        4.2.3 Security Model
    4.3 The Proposed Scheme
        4.3.1 Basic Components
        4.3.2 Main Idea
        4.3.3 The Concrete Construction
        4.3.4 Correctness
    4.4 Security Analysis
    4.5 Performance Analysis
        4.5.1 Efficiency Analysis
        4.5.2 Experimental Evaluation
    4.6 Summary
Chapter 5 Privacy-Preserving Federated Deep Learning Scheme
    5.1 Overview
    5.2 Problem Statement
        5.2.1 System Model
        5.2.2 Definitions of DeepPAR and DeepDPA
    5.3 The Proposed Scheme
        5.3.1 Basic Components
        5.3.2 Security Requirements
        5.3.3 Main Idea
        5.3.4 DeepPAR Based on Re-encryption
        5.3.5 DeepDPA Based on Group Key Management
    5.4 Security Analysis
    5.5 Performance Analysis
        5.5.1 Efficiency Analysis
        5.5.2 Experimental Evaluation
    5.6 Summary
Chapter 6 Privacy-Preserving and Verifiable Online Crowdsourcing Scheme Deployedin Machine Learning
    6.1 Overview
    6.2 Problem Statement
        6.2.1 System Model
        6.2.2 Definitions of PVOC
        6.2.3 Threat Model
        6.2.4 Design Goal
    6.3 The Proposed Scheme
        6.3.1 Basic Components
        6.3.2 Main Idea
        6.3.3 The Concrete Construction
    6.4 Security Analysis
    6.5 Performance Analysis
        6.5.1 Efficiency Analysis
        6.5.2 Experimental Evaluation
    6.6 Summary
Chapter 7 Conclusion and Future Work
    7.1 Conclusion
    7.2 Future Work
Bibliography
Acknowledgement
作者简介



本文编号：3456556