International Association
for Cryptologic Research

A common drawback of secure vector summation protocols in the single-server model is that they impose at least one synchronization point between all clients contributing to the aggregation. This results in clients waiting on each other to advance through the rounds of the protocol, leading to large latency (or failures due to too many dropouts) even if the protocol is computationally efficient. In this paper we propose protocols in the single-server model where clients contributing data to the aggregation (i) send a single message to the server and (ii) can join aggregation sessions dynamically whenever they have resources, i.e., without the need for synchronizing their reporting time with any other clients. Our approach is based on a committee of parties that aid in the computation by running a setup phase before data collection starts, and a verification/decryption phase once it ends. Unlike existing committee-based protocols such as Flamingo (S&P 2023), the cost for committee members can be made sub-linear in the number of clients, and does not depend on the size of the input client vectors. Our experimental evaluation shows that our protocol, even while allowing dynamic client participation, is competitive with the state of the art protocols that do not have that feature in both computation and communication.

In this work we present a new construction for a batched Oblivious Pseudorandom Function (OPRF) based on Vector-OLE and the PaXoS data structure. We then use it in the standard transformation for achieving Private Set Intersection (PSI) from an OPRF. Our overall construction is highly efficient with $O(n)$ communication and computation. We demonstrate that our protocol can achieve malicious security at only a very small overhead compared to the semi-honest variant. For input sizes $n = 2^{20}$, our malicious protocol needs 6.2 seconds and less than 59 MB communication. This corresponds to under 450 bits per element, which is the lowest number for any published PSI protocol (semi-honest or malicious) to date. Moreover, in theory our semi-honest (resp. malicious) protocol can achieve as low as 219 (resp. 260) bits per element for $n=2^{20}$ at the added cost of interpolating a polynomial over $n$ elements. As a second contribution, we present an extension where the output of the PSI is secret-shared between the two parties. This functionality is generally referred to as Circuit-PSI. It allows the parties to perform a subsequent MPC protocol on the secret-shared outputs, e.g., train a machine learning model. Our circuit PSI protocol builds on our OPRF construction along with another application of the PaXoS data structure. It achieves semi-honest security and allows for a highly efficient implementation, up to 3x faster than previous work.