CRII: SaTC: Enabling Secure Machine Learning Queries over Encrypted Database in Cloud Computing

该奖项全部或部分由《2021年美国救援计划法案》（公法117-2）资助。在云计算中，公共云服务提供商可以提供云存储作为主要服务，同时通过使用存储中的客户数据提供额外的基于机器学习（ML）的服务。虽然这种商业模式很有前景，但它也带来了安全问题，因为公共商业云不能完全信任。例如，公共商业云可能将客户的敏感数据出售给第三方。在这个项目中，研究者开发了一个方案，在云存储中的加密数据库上实现安全和隐私保护的机器学习服务。该项目更广泛的意义和重要性是双重的。首先，该项目扩展了云计算服务的边界，带来了新的业务增长可能性，使存储在公共云上的数据能够更安全、更隐私地进行人工智能分析。其次，该项目让女性和少数族裔学生参与计算，从而培养21世纪具有数据能力的劳动力。在这个项目中，研究者开发了一个方案，通过采用索引辅助方法来支持云存储中加密数据库的安全ML查询。在这种方法中，数据集中的每个数据项（即训练示例）使用AES/DES进行正式加密，以实现强密文隐私。对于每个数据项，都会生成一个支持ML分析的安全索引项。因此，该方案同时实现了强大的密文隐私和ML功能，以及索引、模型和令牌的隐私。总之，这个项目做出了两个重要的技术贡献。首先，开发一种索引辅助方法，解决数据安全性与实用性之间的冲突。其次，人工智能驱动的数据加密/掩码技术的发展，在隐私和运行时性能方面优于差分隐私和同态加密。该奖项反映了美国国家科学基金会的法定使命，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。

This award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117-2).In cloud computing, public cloud service providers can provide cloud storage as the primary service, while providing additional machine learning (ML)-based services by using the clients' data in storage. Although this business model is promising, it also brings in security concerns since the public commercial cloud cannot be fully trusted. For example, public commercial clouds may sell clients' sensitive data to third parties. In this project, the investigator develops a scheme to enable secure and privacy-preserving machine learning services over the encrypted database in cloud storage. The project's broader significance and importance are two-fold. First, the project extends the border of cloud computing services and brings in new business growth possibilities that enable safer and privacy-preserving AI analyses on data stored in public clouds. Second, the project engages female and under-represented minority students in computing, thus fostering the 21st-century data-capable workforce. In this project, the investigator develops a scheme to support secure ML queries over encrypted databases in cloud storage by employing an index-aid approach. In this approach, each data item (i.e., a training example) in a dataset is formally encrypted using AES/DES to achieve strong ciphertext privacy. For each data item, a secure index item that can support ML analyses is generated. Therefore, this scheme achieves strong ciphertext privacy and ML capability simultaneously, as well as index, model, and token privacy. In summary, this project makes two significant technical contributions. First, the development of an index-aid approach that addresses the conflict between data security and utility. Second, the development of AI-driven data encryption/mask techniques that outperform differential privacy and homomorphic encryption regarding privacy and runtime performance.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.