Practical Private Regular Expression Matching

Abstract

Abstract. Regular expressions are a frequently used tool to search in large texts. They provide the ability to compare against a structured pattern that can match many text strings and are common to many applications, even programming languages. This paper extends the problem to the private two-party setting where one party has the text string and the other party has the regular expression. The privacy constraint is that neither party should learn about the input of the other party, i.e. the string or the regular expression, except the result of the computation which is wether the string matches the regular expression or not. Secure Multiparty Computation provides general algorithms for any such problem, but it has been recommended to develop special protocols for important cases that provide better performance using the domain knowledge of that problem. This paper presents two protocols: One with perfect secrecy that provides a lower-bound on protocols using circuit construction and a fast one that provides better performance bounds, but the secrecy it provides is limited and tuned for practical applications. The fast protocol presented here uses permutation and commutative encryption as its only building blocks.

Deploying Convolutional Networks on Untrusted Platforms Using 2D Holographic Reduced Representations

“…The second broad class of approaches is protocol-based, requiring multiple rounds of communication where data is sent back-and-forth between the host that is requesting computation, and the third party server performing the bulk of computation. Methods like Secure Multi-party Com-putation (SMC) (Kerschbaum, 2006;Du & Atallah, 2001) and other "protocols" are developed on top of "Oblivious Transfer" (OT), a primitive by which a sender and receiver exchange messages (Rabin, 1981). Many OT protocols 3 have been customized for deep learning applications (Riazi et al, 2018;Rouhani et al, 2018;Chandran et al, 2019;Riazi et al, 2019;Liu et al, 2017;Mohassel & Zhang, 2017), but suffer similar limitations to FHE.…”

“…The current solutions to this situation naturally come from the encryption community, and the tools of Secure Multiparty Computation (SMC) (Kerschbaum, 2006;Du & Atallah, 2001) and Homomorphic Encryption (HE) (Gilad-Bachrach et al, 2016) provide methods for running programs on untrusted hardware that guarantee the privacy of the results. These are valuable tools, but computationally demanding and limiting.…”