Private Stateful Information Retrieval

Abstract

Private information retrieval (PIR) is a fundamental tool for preserving query privacy when accessing outsourced data. All previous PIR constructions have significant costs preventing widespread use. In this work, we present private stateful information retrieval (PSIR), an extension of PIR, allowing clients to be stateful and maintain information between multiple queries. Our design of the PSIR primitive maintains three important properties of PIR: multiple clients may simultaneously query without complex concurrency primitives, query privacy should be maintained if the server colludes with other clients, and new clients should be able to enroll into the system by exclusively interacting with the server. We present a PSIR framework that reduces an online query to performing one single-server PIR on a sub-linear number of database records. All other operations beyond the single-server PIR consist of cryptographic hashes or plaintext operations. In practice, the dominating costs of resources occur due to the public-key operations involved with PIR. By reducing the input database to PIR, we are able to limit expensive computation and avoid transmitting large ciphertexts. We show that various instantiations of PSIR reduce server CPU by up to 10x and online network costs by up to 10x over the previous best PIR construction.

Advances and Open Problems in Federated Learning

“…Additionally, it has been shown how to construct communication-efficient PIR on a single-server by leveraging side information available to the user [251], for example via client local state. Patel et al [372] presented and implemented a practical hybrid (computational and information theoretic) PIR scheme on a single server assuming client state. Corrigan-Gibbs and Kogan [131] present theoretical constructions for PIR with sublinear online time by working in an offline/online model where, during an offline phase, clients fetch information from the server(s) independent on the future query to be performed.…”

ConsenSGX: Scaling Anonymous Communications Networks with Trusted Execution Environments

“…For instance, today the Tor network has approximately 800 exits and the remaining 5500 serve as guards and middles; scaling that up to a total of 100,000 relays while preserving the ratio of exit to middle relays, implies a query to a pool with 13,000 exit relays, which takes 472.2 ms ± 0.4 ms end-to-end time assuming a query with B = 50, and another to a pool with 87,000 middle relays, which takes 517.8 ms ± 0.5 ms, resulting in a total time of 990 ms ± 0.6 ms, or barely one second. In our experiments we use XPIR as the benchmark for CPIR, but we note that there have been recent advancements in CPIR constructions since XPIR, namely SealPIR [4] and PSIR [54]. Although we do not perform empirical comparisons against these two schemes in this work, here we provide intuitions as to why these schemes cannot enable scalability to the extent that ConsenSGX does.…”

Survey on Federated Learning Towards Privacy Preserving AI

“…Research works propose to leverage local client state to speed up PIR. Patel et al [91] showed a practical hybrid PIR scheme on a single server was implemented and validated. Corrigan-Gibbs and Kogan [92] present protocols for PIR with sublinear online time by working in an offline/online model.…”