SaTC: TTP: Small: Modular Platform for Web-based Secure Multi-Party Analytics

该项目设计，开发和应用模块化基础设施，用于构建基于Web的应用程序，使个人和组织能够在数据共享受到保密问题，法律的限制或公司政策阻碍的情况下从隐私保护数据聚合和分析中受益。今天，个人和组织面临着可以收集和处理的有价值数据的爆炸与由于恶意行为者，犯罪企业和软件错误而暴露数据（可能是敏感的）的威胁之间的紧张关系。作为回应，各实体往往隔离其数据，并在此过程中放弃从协作数据分析中获益的机会。该基础设施通过允许软件开发人员、企业家、社会科学家和组织构建基于Web的应用程序来解决这些问题，这些应用程序利用了安全多方计算（MPC），这是一种已经存在了35年的加密技术。在过去的十年中，MPC技术已经在几个软件框架中实现，并且MPC的专业化个人部署包括税务欺诈检测，疾病监测和薪酬公平评估。然而，MPC的社会效益只有在能够设计和组装轻量级、用户友好、基于网络的MPC应用程序时才能广泛实现，这些应用程序可以让没有网络安全背景的决策者和利益相关者理解，也可以让公众和服务不足的人群访问和使用。在本项目过程中开发和评估的软件库、软件包和应用程序将对多个人、组织和政策制定者使用敏感数据语料库的方式产生切实影响，以确定趋势、诊断问题、测试假设和为政策决策提供信息。本项目降低了设计、开发、通过提供两种类型的开源软件来部署MPC应用：（1）用于构建后端框架和应用程序编程接口（API）的库，这些框架和API可以支持基于Web的MPC使能服务，以及（2）用于开发客户端的前端框架，可以在标准浏览器或标准移动终端上运行的附加功能。总的来说，该软件基础设施支持适合各种部署场景和用户角色的MPC功能的多样化集合。这项工作选择、调整、转换、重构、优化和封装现有的MPC算法和框架，重点关注两个运营指标：（1）增强MPC的可访问性并推动其采用，以及（2）在低性能生产环境中以最小的开销运行。MPC技术的早期采用者确定的三个真实用例将为这项工作提供信息和基础：波士顿市和波士顿妇女劳动力理事会的薪酬公平分析，用于成瘾康复的移动的健康干预应用程序，以及根据各种不同组织维护的数据集设计服务不足人群的生计评估。所有三个测试用例都提供了重要的反馈，这些反馈可以指导工作，从而在实践中成功交付和使用MPC，方法是阐明现有MPC功能对此类应用程序的适用性，针对各个场景优化或定制此类功能的必要性，以及围绕真实世界参与者角色和工作流程组织和分解此类功能。

This project designs, develops, and applies a modular infrastructure for building web-based applications that allow individuals and organizations to benefit from privacy-preserving data aggregation and analysis in contexts where data sharing is encumbered by confidentiality concerns, legal restrictions, or corporate policies. Today, individuals and organizations face a tension between the explosion of valuable data that can be collected and processed and the threat of the exposure of data (which may be sensitive) due to malicious actors, criminal enterprises, and software errors. In response, entities often isolate their data, and in the process forego opportunities to benefit from collaborative data analysis. The infrastructure addresses these circumstances by allowing software developers, entrepreneurs, social scientists, and organizations to build web-based applications that leverage Secure Multi-Party Computation (MPC), a collection of cryptographic techniques that have been known for 35 years. In the last decade, MPC techniques have been implemented in several software frameworks, and specialized individual deployments of MPC include work on tax fraud detection, disease surveillance, and pay equity assessment. However, MPC's social benefits are only broadly realized when it is possible to design and assemble lightweight, user-friendly, web-based MPC applications that decision-makers and stakeholders without a cybersecurity background can understand and that the public and underserved populations can access and utilize. Software libraries, packages, and applications developed and evaluated over the course of this project will have tangible impacts on the ways in which sensitive data corpora can be used by multiple individuals, organizations, and policymakers to identify trends, diagnose problems, test hypotheses, and inform policy decisions.This project lowers the barrier for the design, development, and deployment of MPC applications by delivering two types of open-source software: (1) libraries for building back-end frameworks and Application Programming Interfaces (APIs) that can support web-based MPC-enabling services and (2) front-end frameworks for developing client-side functionalities that can operate in a standard browser or on a standard mobile device. Collectively, this software infrastructure enables a diverse collection of MPC functionalities suitable for a variety of deployment scenarios and user roles. The effort selects, adapts, translates, refactors, optimizes, and encapsulates existing MPC algorithms and frameworks with a focus on two operational metrics: (1) enhancing accessibility and driving adoption of MPC, and (2) operating with minimal overhead in low-performance production environments. Three real-world use cases identified by early adopters of MPC technology will inform and ground the work: a pay equity analysis by the City of Boston and the Boston Women's Workforce Council, a mobile health intervention app for addiction recovery, and an effort to design a livelihood assessment of underserved populations based on data sets maintained by a variety of distinct organizations. All three test cases provide essential feedback that can direct efforts leading to successful delivery and utilization of MPC in practice by elucidating the suitability of existing MPC functionalities for such applications, the necessity for optimization or customization of such functionalities for individual scenarios, and the organization and decomposition of such functionalities around real-world participant roles and workflows.