基于区块链的联邦学习:模型、方法与应用
Blockchain-enabled Federated Learning:Models,Methods and Applications
李程 1袁勇 2郑志勇 2杨东 3王飞跃4
作者信息
- 1. 中国人民大学数学学院 北京 100872;中国人民大学交叉科学研究院 北京 100872
- 2. 中国人民大学数学学院 北京 100872
- 3. 中国人民大学交叉科学研究院 北京 100872
- 4. 中国科学院自动化研究所复杂系统管理与控制国家重点实验室 北京 100190;澳门科技大学系统工程研究所 澳门 999078
- 折叠
摘要
近年来,人类社会快速步入大数据时代,数据安全与隐私保护已成为发展大数据生态及相关数字经济的关键问题.联邦学习(Federated learning)作为分布式机器学习的一种新范式,致力于在保护数据隐私的同时从分布式本地数据集中训练全局模型,因而获得了广泛和深入的研究.然而,联邦学习体系面临的中心化架构、激励机制设计和系统安全等技术挑战仍有待进一步研究,而区块链被认为是应对这些挑战的有效解决方案,并已成功应用于联邦学习的许多研究和实践场景.在系统性地梳理现阶段区块链与联邦学习集成研究成果的基础上,提出基于区块链的联邦学习(Blockchain-enabled federated learning,BeFL)概念模型,阐述其中的若干关键技术、研究问题与当前研究进展,探讨该领域的应用场景以及有待进一步研究的关键问题,并讨论未来发展的潜在方向,致力于为构建去中心化和安全可信的数据生态基础设施、促进数字经济与相关产业的发展提供有益的参考与借鉴.
Abstract
In recent years,human society has been witnessed to evolve fast to the era of big data,rendering the data security and privacy protection a key issue for the development of digital economies.Federated learning,as a novel pattern for distributed machine learning,is aimed to train a centralized model from decentralized datasets while protecting user privacy,and is now intensively studied in literature.However,a variety of technical chal-lenges,e.g.,centralized architecture,incentive mechanism design,and system-wide security issues,are still awaiting further research efforts.In this respect,blockchain proves to be an elegant solution for federated learning to over-come these issues,and thus has been applied in federated learning in many scenarios with success.In this paper,we proposed the conceptual model for blockchain-enabled federated learning(BeFL)based on a comprehensive review of related literatures,and discussed the key techniques,research issues,as well as the state-of-the-art research pro-gresses.We also investigated potential application scenarios,several key issues to be addressed and the future trends.Our work is aimed at offering useful reference and guidance for establishing a new infrastructure for decent-ralized,secured and trusted data ecosystem,and also promoting the development of digital economy industries.
关键词
区块链/联邦学习/智能合约/机器学习/隐私保护Key words
Blockchain/federated learning/smart contract/machine learning/privacy protection引用本文复制引用
基金项目
国家自然科学基金(72171230)
澳门科学技术发展基金(0050/2020/A1)
北京市未来区块链与隐私计算高精尖创新中心项目()
出版年
2024
国家科技期刊平台
NSTL
万方数据