Private Aggregation from Fewer Anonymous Messages

Abstract

Consider the setup where n parties are each given an element in the finite field and the goal is to compute the sum in a secure fashion and with as little communication as possible. We study this problem in the anonymized model of Ishai et al. (FOCS 2006) where each party may broadcast anonymous messages on an insecure channel. We present a new analysis of the one-round “split and mix” protocol of Ishai et al. In order to achieve the same security parameter, our analysis reduces the required number of messages by a multiplicative factor. We also prove lower bounds showing that the dependence of the number of messages on the domain size, the number of parties, and the security parameter is essentially tight. Using a reduction of Balle et al. (2019), our improved analysis of the protocol of Ishai et al. yields, in the same model, an -differentially private protocol for aggregation that, for any constant and any , incurs only a constant error and requires only a constant number of messages per party. Previously, such a protocol was known only for messages per party.

Advances and Open Problems in Federated Learning

“…More generally, shuffling models of differential privacy can use broader classes of local randomizers, and can even select these local randomizers adaptively [178]. This can enable differentially private protocols with far smaller error than what is possible in the local model, while relying on weaker trust assumptions than in the central model, e.g., [120,178,45,201,204,200,202,203,110].…”

Aggregation and Transformation of Vector-Valued Messages in the Shuffle Model of Differential Privacy

“…Also relevant to our research is the work of Ghazi et al [22], which explored the related problems of private frequency estimation and selection in a similar context, drawing comparisons between the errors achieved in the Single-Message Shuffle Model and the Multi-Message Shuffle Model. A similar team of authors produced a follow-up paper [23] describing a more efficient protocol for private summation in the Single-Message Shuffle Model, using the 'invisibility cloak' technique to facilitate the addition of zero-sum noise without coordination between the users. The most recent work of Ghazi et al [24] relaxes the single-message requirement of their previous protocols to improve the accuracy of private summation in the Shuffle Model to be close to that of the Centralized Model.…”

What do information centric networks, trusted execution environments, and digital watermarking have to do with privacy, the data economy, and their future?

“…For the smooth integration of PD-ICN with the existing Web and end users the "Interface with Existing Web" module and "User Privacy Enhancements" module are responsible to interact with the existing monetization model of the Web and enhance end users privacy, respectively. Finally, the "Data Access Environment (APIs)" acts as a direct link between the backend interface of the Data Layer and the Application Layer Open Questions: Different approaches can be found in the literature pertaining anonymous data analysis though aggregation techniques [16,27,60] and Secure, Privacy Preserving Machine Learning (SPP-ML) [4,34,36,46], focusing on different aspects of SPP-ML such as ML as a Service (MLaaS) [36], ML with multiple data providers [46] and verifiable outsourcing schemes [34]. The open questions under the suggested PD-ICN setting are: (1) How to develop a "sandboxed" environment (or combine existing solutions) to execute ML algorithms on top of PD-ICN data at the application layer?…”