Blockchain-orchestrated machine learning for privacy preserving federated learning in electronic health data

【Author】 Passerat-Palmbach, Jonathan; Farnan, Tyler; McCoy, Mike; Harris, Justin D.; Manion, Sean T.; Flannery, Heather Leigh; Gleim, Bill

【Source】2020 IEEE INTERNATIONAL CONFERENCE ON BLOCKCHAIN (BLOCKCHAIN 2020)

【Abstract】Machine learning and blockchain technology have been explored for potential applications in medicine with only modest success to date. Focus has shifted to exploring the intersection of these technologies along with other privacy preserving encryption techniques for better utility. This combination applied to federated learning, which allows remote execution of function and analysis without the need to move highly regulated personal health information, seems to be the key to successful applications of these technologies to rapidly advance evidence-based medicine. We give a brief history of these technologies in medicine, outlining some of the challenges with successful use. We then explore a more detailed combination of usage with an emphasis on decentralizing or federating the learning process along with auditability and incentivization blockchain can allow in the machine learning process. Based on the cost-benefit analysis of previous efforts, we provide the framework for an advanced blockchain-orchestrated machine learning system for privacy preserving federated learning in medicine and a new utility in health. Six critical elements for this approach in the future will be: (a) Data and analytic processes discoverable on secure public blockchain while retaining privacy of the data and analytic processes (b) Value fabricated by generating data/compute matches that were previously illegal, unethical and infeasible (c) Compute guarantees provided by federated learning and advanced cryptography (d) Privacy guarantees provided by software (e.g., Homomorphic Encryption, Secure Multi-Party Computation, ...) and hardware (e.g., Intel SGX and AMD SEV-SNP) cryptography (e) Data quality incentivized via tokenized reputation-based rewards (f) Discarding of poor data accomplished via model poisoning attack prevention techniques.

【Keywords】

【标题】区块链编排机器学习，用于电子健康数据中的隐私保护联邦学习

【摘要】机器学习和区块链技术已被探索用于医学中的潜在应用，迄今为止仅取得了适度的成功。重点已转移到探索这些技术与其他隐私保护加密技术的交叉点，以获得更好的实用性。这种组合应用于联邦学习，允许远程执行功能和分析，而无需移动高度监管的个人健康信息，似乎是成功应用这些技术以快速推进循证医学的关键。我们简要介绍了这些技术在医学中的历史，概述了成功使用的一些挑战。然后，我们探索了更详细的使用组合，重点是分散或联邦学习过程，以及区块链可以在机器学习过程中允许的可审计性和激励。基于对先前工作的成本效益分析，我们为先进的区块链编排机器学习系统提供了框架，用于保护隐私的医学联邦学习和健康的新实用程序。未来这种方法的六个关键要素将是：（a）可在安全的公共区块链上发现数据和分析过程，同时保留数据和分析过程的隐私（b）通过生成以前非法、不道德的数据/计算匹配来制造价值(c) 联邦学习和高级密码学提供的计算保证 (d) 软件（例如同态加密、安全多方计算……）和硬件（例如 Intel SGX 和 AMD SEV-SNP）提供的隐私保证) 密码学 (e) 通过标记化的基于声誉的奖励来激励数据质量 (f) 通过模型中毒攻击预防技术丢弃不良数据。

【关键词】无

【发表时间】2020

【收录时间】2022-07-06

【文献类型】Proceedings Paper

【论文大主题】区块链联邦学习

【论文小主题】联邦学习为主体

【翻译者】石东瑛