SBIR Phase I: Accelerating Machine Learning on Encrypted Data

这个小型企业创新研究（SBIR）第一阶段项目的更广泛影响将是使组织能够在不泄露私人或专有信息的情况下从数据中协作和提取见解。数据驱动的创新使我们生活的各个方面受益，从更好的医疗保健和提高生产力到更智能的能源消耗。人工智能（AI）具有巨大的潜力，但数据访问已成为主要瓶颈。该提案所实现的改进的数据安全和隐私技术将消除实现这些好处的障碍。该技术的商业化将对云计算和数据分析市场产生重大影响。该技术消除了阻止公司将计算转移到云的安全风险，云提供了负担得起且可扩展的数据处理基础设施。此外，组织可以在不泄露底层信息的情况下对敏感数据进行协作。这个小型企业创新研究第一阶段项目将基于全同态加密（FHE）对加密数据进行机器学习和人工智能。目前的方法太慢，难以使用;因此，它们在企业环境中的适用性有限。这个第一阶段项目通过设计一个硬件加速的FHE服务来扩展当前的方法，该服务符合强制性的安全标准并针对企业AI应用程序。为了快速推进翻译，将使用现成的加速器，这些加速器可在主要的云计算数据中心使用。进一步的性能提升将通过减少加密计算的计算开销的神经网络优化来实现。除了性能改进之外，还将实施与通用机器学习框架的软件集成，以降低可用性障碍。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact of this Small Business Innovation Research (SBIR) Phase I project will be to enable organizations to collaborate and extract insights from data without revealing private or proprietary information. Data-driven innovations benefit all aspects of our lives, ranging from better healthcare and increased productivity to smarter energy consumption. Artificial intelligence (AI) has so much potential but access to data has become the major bottleneck. Improved data security and privacy techniques enabled by this proposal will remove barriers to realizing these benefits. Commercialization of the proposed technology will have major implications for the cloud computing and data analytics market. The proposed technology eliminates the security risks that prevent companies from moving computations to the cloud, which offers affordable and scalable data processing infrastructure. Furthermore, organizations can collaborate on sensitive data without revealing the underlying information.This Small Business Innovation Research Phase I project will enable machine learning and AI on encrypted data based on Fully-Homomorphic Encryption (FHE). Current approaches are too slow and difficult to use; therefore, they have limited applicability in enterprise settings. This Phase I project extends current approaches by designing a hardware-accelerated FHE service that meets mandated security standards and targets enterprise AI applications. To advance translation quickly, off-the-shelf accelerators will be used, which are available in major cloud computing data centers. Further performance gains will be achieved through neural network optimizations that reduce the computation overhead of encrypted computations. In addition to the performance improvements, software integration with common machine learning frameworks will be implemented to lower the barrier for usability.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.