Practically Efficient Protocols for Private Computation using Homomorphic Encryption

Abstract

Digital services have become an indispensable part of our daily lives, particularly services that interact with our most private and sensitive data. With the abundance of such services, users are left to make the difficult choice: can I safely use digital services and products, or does it necessarily come at the cost of my privacy. Private computation techniques empower service providers to perform computation over private data, without the need to observe the data. This not only provides privacy for clients while the data is being used but reduces the risk of incidents such as data leaks for service providers. One commonly used tool for private computation is Homomorphic Encryption (HE), which is a form of encryption that allows computation on data in encrypted form. While homomorphic encryption in theory permits arbitrary computation over encrypted data, in practice, a naive implementation of a desired functionality rarely yields a practical result. For example, one common obstacle when using homomorphic encryption is the high computation time and the large ciphertexts that incur high network costs. However, communication and computation costs are not the only metrics that need to be considered. In my work, we describe problems that arise when homomorphic encryption is used in applications and address these limitations by proposing new techniques and novel protocols. In these new constructions, we not only improve the performance compared to prior work in terms of communication and computation costs but also address additional problems that arise in the deployment of these protocols. Throughout the process, we draw insights on how to design protocols that can be applicable for developers, practitioners, and future researchers. For example, we enable homomorphic comparison of encrypted numbers with higher precision than previous work, using novel representation of numbers that is more suitable for homomorphic encryption. Using this and other building blocks, we propose efficient protocols for decision tree evaluation and private set intersection. Moreover, through our work on private information retrieval, we identify the challenges of using such a protocol in practice and propose novel protocols that are suited for deployment in real-world applications.