Towards Personalized Differentially Private Learning for Decentralized Local Graphs
2026-07-06 • Machine Learning
Machine LearningCryptography and Security
AI summaryⓘ
The authors address the problem of protecting user privacy when collecting graph data that is spread out across many devices or users. They point out that existing methods treat everyone's privacy needs the same, which leads to too much noise and less useful data. Their solution, called PPGNN, allows each user to set their own level of privacy, which helps keep the data more accurate while still protecting individuals. The approach uses two steps: one to add personalized noise and another to adjust the data to reduce distortion. Tests on real-world data show their method balances privacy and usefulness well.
Graph-structured dataDecentralized dataLocal Differential Privacy (LDP)Personalized privacyNoise injectionData perturbationPrivacy budgetGraph learningData utilityPrivacy-preserving methods
Authors
Longzhu He, Peng Tang, Chaozhuo Li, Jinhu Fu, Litian Zhang, Li Sun, Philip S. Yu, Sen Su
Abstract
Graph-structured data is increasingly generated and stored in decentralized environments, such as social platforms, mobile applications, and edge networks, where users maintain control over their local graph data. However, collecting and analyzing such decentralized graph data for downstream learning tasks raises significant privacy concerns, as nodes and their attributes often contain sensitive personal information. Local Differential Privacy (LDP) has emerged as a promising solution for privacy-preserving data collection without relying on trusted servers. Nevertheless, existing LDP-based graph learning methods typically assume uniform privacy requirements across users, ignoring the heterogeneous and personalized privacy preferences commonly observed in real-world systems. This uniform treatment leads to inflexible noise injection at the data collection stage, resulting in substantial distortion of graph data and degraded utility in subsequent analysis. To address this limitation, we propose PPGNN, a personalized differentially private framework for decentralized graph data. PPGNN enables user-specific privacy budgets during local perturbation while preserving analytical utility. To handle heterogeneous privacy levels and noise distortion, we design a two-stage solution consisting of a Personalized Perturbation Mechanism (PPM) and a weighted calibration strategy, FlexProp. Extensive experiments on six real-world graph datasets demonstrate that PPGNN effectively balances personalized privacy protection and data utility in decentralized graph learning scenarios.