Muffliato: Peer-to-Peer Privacy Amplification for Decentralized Optimization and Averaging

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Cyffers, Edwige, Even, Mathieu, Bellet, Aurélien, Massoulié, Laurent
Format: Preprint
Published: 2022
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910480209018880
author Cyffers, Edwige
Even, Mathieu
Bellet, Aurélien
Massoulié, Laurent
author_facet Cyffers, Edwige
Even, Mathieu
Bellet, Aurélien
Massoulié, Laurent
contents Decentralized optimization is increasingly popular in machine learning for its scalability and efficiency. Intuitively, it should also provide better privacy guarantees, as nodes only observe the messages sent by their neighbors in the network graph. But formalizing and quantifying this gain is challenging: existing results are typically limited to Local Differential Privacy (LDP) guarantees that overlook the advantages of decentralization. In this work, we introduce pairwise network differential privacy, a relaxation of LDP that captures the fact that the privacy leakage from a node $u$ to a node $v$ may depend on their relative position in the graph. We then analyze the combination of local noise injection with (simple or randomized) gossip averaging protocols on fixed and random communication graphs. We also derive a differentially private decentralized optimization algorithm that alternates between local gradient descent steps and gossip averaging. Our results show that our algorithms amplify privacy guarantees as a function of the distance between nodes in the graph, matching the privacy-utility trade-off of the trusted curator, up to factors that explicitly depend on the graph topology. Finally, we illustrate our privacy gains with experiments on synthetic and real-world datasets.
format Preprint
id arxiv_https___arxiv_org_abs_2206_05091
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle Muffliato: Peer-to-Peer Privacy Amplification for Decentralized Optimization and Averaging
Cyffers, Edwige
Even, Mathieu
Bellet, Aurélien
Massoulié, Laurent
Cryptography and Security
Distributed, Parallel, and Cluster Computing
Machine Learning
Decentralized optimization is increasingly popular in machine learning for its scalability and efficiency. Intuitively, it should also provide better privacy guarantees, as nodes only observe the messages sent by their neighbors in the network graph. But formalizing and quantifying this gain is challenging: existing results are typically limited to Local Differential Privacy (LDP) guarantees that overlook the advantages of decentralization. In this work, we introduce pairwise network differential privacy, a relaxation of LDP that captures the fact that the privacy leakage from a node $u$ to a node $v$ may depend on their relative position in the graph. We then analyze the combination of local noise injection with (simple or randomized) gossip averaging protocols on fixed and random communication graphs. We also derive a differentially private decentralized optimization algorithm that alternates between local gradient descent steps and gossip averaging. Our results show that our algorithms amplify privacy guarantees as a function of the distance between nodes in the graph, matching the privacy-utility trade-off of the trusted curator, up to factors that explicitly depend on the graph topology. Finally, we illustrate our privacy gains with experiments on synthetic and real-world datasets.
title Muffliato: Peer-to-Peer Privacy Amplification for Decentralized Optimization and Averaging
topic Cryptography and Security
Distributed, Parallel, and Cluster Computing
Machine Learning
url https://arxiv.org/abs/2206.05091