Honeycrisp

Abstract

Recently, a number of systems have been deployed that gather sensitive statistics from user devices while giving differential privacy guarantees. One prominent example is the component in Apple's macOS and iOS devices that collects information about emoji usage and new words. However, these systems have been criticized for making unrealistic assumptions, e.g., by creating a very high "privacy budget" for answering queries, and by replenishing this budget every day, which results in a high worst-case privacy loss. However, it is not obvious whether such assumptions can be avoided if one requires a strong threat model and wishes to collect data periodically, instead of just once. In this paper, we show that, essentially, it is possible to have one's cake and eat it too. We describe a system called Honeycrisp whose privacy cost depends on how often the data changes, and not on how often a query is asked. Thus, if the data is relatively stable (as is likely the case, e.g., with emoji and word usage), Honeycrisp can answer periodic queries for many years, as long as the underlying data does not change too often. Honeycrisp accomplishes this by using a) the sparsevector technique, and b) a combination of cryptographic techniques to enable global differential privacy without a trusted party. Using a prototype implementation, we show that Honeycrisp is efficient and can scale to large deployments.

OLYMPIA: A Simulation Framework for Evaluating the Concrete Scalability of Secure Aggregation Protocols

“…For secure aggregation protocols, authors typically do not report end-to-end wall-clock time resulting from an experimental evaluation, because it is not feasible to run such an experiment [3,5,15,17,22,23,29,30,33]. As we discuss later, the inability to measure concrete performance of these protocols makes it difficult to understand their relative performance properties.…”

“…Several systems have been developed for differentially private analytics (i.e. database queries) that leverage ideas from secure aggregation, including Honeycrisp [22], Orchard [23], and Cryptε [25]. These systems are designed to scale to millions of participants, using specialized protocols and a slightly weaker threat model.…”

“…Due to its scale, the later Bell et al protocol [3] included concrete results only for the number of neighbors each client must communicate with, and did not include an experimental evaluation of computation or communication cost. Similarly, Honeycrisp [22] and Orchard [23] include concrete experimental results for parts of their protocol, but do not perform end-to-end experiments due to the protocol's intended scale.…”

Federated Learning of Artificial Neural Networks

“…That means no one, including the server or some trusted entities, should be able to make "very concrete" conclusions about the data, what the user owns. Analogously, when the communication involves secret key protocols, the most appropriate approach would be holding it at the clients, that is, to share among parties as it has been proposed by [176] and [181].…”