Shuffled Check-in: Privacy Amplification towards Practical Distributed Learning

Abstract

Recent studies of distributed computation with formal privacy guarantees, such as differentially private (DP) federated learning, leverage random sampling of clients in each round (privacy amplification by subsampling) to achieve satisfactory levels of privacy. Achieving this however requires strong assumptions which may not hold in practice, including precise and uniform subsampling of clients, and a highly trusted aggregator to process clients' data. In this paper, we explore a more practical protocol, shuffled check-in, to resolve the aforementioned issues. The protocol relies on client making independent and random decision to participate in the computation, freeing the requirement of server-initiated subsampling, and enabling robust modelling of client dropouts. Moreover, a weaker trust model known as the shuffle model is employed instead of using a trusted aggregator. To this end, we introduce new tools to characterize the Rényi differential privacy (RDP) of shuffled check-in. We show that our new techniques improve at least three times in privacy guarantee over those using approximate DP's strong composition at various parameter regimes. Furthermore, we provide a numerical approach to track the privacy of generic shuffled check-in mechanism including distributed stochastic gradient descent (SGD) with Gaussian mechanism. To the best of our knowledge, this is also the first evaluation of Gaussian mechanism within the local/shuffle model under the distributed setting in the literature, which can be of independent interest.

Federated learning and differential privacy in clinical health: Extensive survey

“…Based on Federated Learning [118] Proposed dynamic secrete sharing mechanism for privacy protection Reduce time overhead through elliptic curve cryptosystem Only considers time overhead for efficiency while ignoring other metrics such as data distribution, false positives and negatives, recall etc Secrete sharing, two-mask protocol homogeneous linear recursive equation homomorphic hash function [98] elliptic curve crypto-system [119] full dynamic secret sharing [120] Framework Shuffled Check-in: Privacy Amplification towards Practical Distributed Learning [121] proposed a protocol for privacy using numerical evaluation of Gaussian mechanism Did not cover fairness with regards to biased data SGD using Gaussian mechanism Rényi differential privacy (RDP) Framework On Privacy and Personalization in Cross-Silo Federated Learning [122] Provided an empirical and theoretical study on meanregularized multi-task learning (MR-MTL) as effective model personalization Understands how privacy in device-silo differs from that in cross-silo FL MR-MTL DP-SGD for noise reduction Framework HealthCare EHR: A Blockchain-Based Decentralized Application [123] Use Ethereum blockchain to build a peer to peer network platform for distributed database of health entities This was to resolve a challenge of data lake that centrally puts the Payment transaction through banks have not been institutionalized…”