PrivGNN: High-Performance Secure Inference for Cryptographic Graph Neural Networks

2025-11-04

AI Summary

To address the lack of privacy protection for graph-structured data in secure graph neural network (GNN) inference under cloud environments, this paper proposes PrivGNN—the first efficient two-party secure inference framework tailored for GNNs. PrivGNN innovatively integrates additive secret sharing (ASS) and function secret sharing (FSS) to design lightweight, interactive protocols supporting both linear and nonlinear GNN layers, achieving a favorable trade-off between security and efficiency. Unlike existing secure inference (SI) schemes primarily designed for images or text, PrivGNN is the first to systematically tackle the structural sensitivity and high computational overhead inherent in encrypted inference over graph-structured (GS) data. Extensive experiments on four benchmark datasets demonstrate that PrivGNN accelerates inference by 1.3–4.7× over state-of-the-art baselines, with negligible accuracy degradation (<0.5%), thereby significantly outperforming existing approaches.

Technology Category

• Machine Learning: Privacy
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Natural Language Processing: Safety and Robustness

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Security and Privacy: Privacy-enhancing technologies
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Graph neural networks (GNNs) are powerful tools for analyzing and learning from graph-structured (GS) data, facilitating a wide range of services. Deploying such services in privacy-critical cloud environments necessitates the development of secure inference (SI) protocols that safeguard sensitive GS data. However, existing SI solutions largely focus on convolutional models for image and text data, leaving the challenge of securing GNNs and GS data relatively underexplored. In this work, we design, implement, and evaluate $sysname$, a lightweight cryptographic scheme for graph-centric inference in the cloud. By hybridizing additive and function secret sharings within secure two-party computation (2PC), $sysname$ is carefully designed based on a series of novel 2PC interactive protocols that achieve $1.5 imes sim 1.7 imes$ speedups for linear layers and $2 imes sim 15 imes$ for non-linear layers over state-of-the-art (SotA) solutions. A thorough theoretical analysis is provided to prove $sysname$'s correctness, security, and lightweight nature. Extensive experiments across four datasets demonstrate $sysname$'s superior efficiency with $1.3 imes sim 4.7 imes$ faster secure predictions while maintaining accuracy comparable to plaintext graph property inference.

Problem

• Securing graph neural network inference in cloud environments
• Addressing privacy gaps in secure inference for graph-structured data
• Developing lightweight cryptographic protocols for graph-centric computations

Innovation

• Hybrid additive and function secret sharing
• Novel two-party computation interactive protocols
• Lightweight cryptographic scheme for graph inference