TGHE: Template-based Graph Homomorphic Encryption for Privacy-Preserving GNN Inference in Edge-Cloud Systems

2026-06-25 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy · Depth: Expert, quick

Summary

TGHE (Template-based Graph Homomorphic Encryption) is a novel ego-centric framework designed for privacy-preserving Graph Neural Network (GNN) inference in edge-cloud systems. It addresses the limitations of existing graph-centric homomorphic encryption (HE)-based GNN systems, which tie per-query cost to global graph size and are restricted to graphs of at most ~20k nodes, making them impractical for large-scale dynamic financial graphs. TGHE resolves this by leveraging a "template phenomenon" where local computation trees in transaction graphs converge into a small set of structural shapes. The framework canonicalizes ego-graphs at the edge and packs structurally identical trees into shared CKKS ciphertexts, enabling SIMD-parallel encrypted inference. It incorporates two optimizers, Approximate Template Fitting and Topology Collapse, to ensure comprehensive SIMD coverage. On the DGraphFin dataset, which contains 3.7M nodes and 4.3M edges, TGHE-Collapse achieved a 66.9x speedup compared to the sequential encrypted baseline, with a minimal AUC loss of less than 0.002.

Key takeaway

For AI Security Engineers designing privacy-preserving GNN systems for large-scale dynamic graphs, TGHE offers a critical paradigm shift. You should consider adopting an ego-centric, template-based approach to overcome the scalability limitations of traditional graph-centric homomorphic encryption. This enables efficient SIMD-parallel encrypted inference on millions of nodes, significantly boosting performance and preserving privacy for sensitive data like financial transactions.

Key insights

Ego-centric graph processing with template-based packing enables efficient, privacy-preserving GNN inference on large dynamic graphs.

Principles

• Graph computation trees often share structural templates.
• Packing identical structures enables SIMD-parallel HE.
• Ego-centric processing scales better for large graphs.

Method

TGHE canonicalizes ego-graphs at the edge, then packs structurally identical trees into shared CKKS ciphertexts for SIMD-parallel encrypted inference, using Approximate Template Fitting and Topology Collapse for full SIMD coverage.

In practice

• Use ego-centric processing for large dynamic graphs.
• Identify structural templates in graph computations.
• Apply SIMD-parallel HE for encrypted inference.

Topics

• Homomorphic Encryption
• Graph Neural Networks
• Privacy-Preserving AI
• Edge-Cloud Systems
• SIMD Parallelism
• Financial Graphs
• CKKS Ciphertext

Best for: Research Scientist, AI Architect, AI Engineer, AI Scientist, Machine Learning Engineer, AI Security Engineer

Related on AIssential

• Beyond Homophily: Towards Generalized Graph Reconstruction Attack and Defense — GNNs are vulnerable to graph reconstruction attacks, but a Markov chain approximation can inform both stronger attacks and effective defenses.
• POLAR-Bench: A Diagnostic Benchmark for Privacy-Utility Trade-offs in LLM Agents — POLAR-Bench diagnoses LLM agent privacy-utility trade-offs, showing frontier models significantly outperform smaller, on-device counterparts.
• TCHG: Tri-Trust Conditioned Heterogeneous Graph Learning for Reliable Dynamic Trust Prediction — Trust prediction benefits from decomposing heterogeneous evidence into functionally distinct channels for graph propagation.
• LSEnet: A Smarter Way to Organize Data Using Curved Space — LSEnet uses hyperbolic space and differentiable structural information for gradient-based, parameter-free deep graph clustering.
• Beyond Soft Masks: Hard-Perturbation Mixup Explainer for Robust GNN Explainability — HPME uses hard perturbations and structural mixup to generate robust, in-distribution GNN explanations by eliminating irrelevant information.
• Giskard : Byzantine Robust and Confidential Aggregation for Large-Scale Decentralized Learning — Giskard enables scalable, confidential, and Byzantine-robust decentralized learning aggregation using a tree-based committee structure and MPC.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.