Search papers, labs, and topics across Lattice.
This paper introduces TGHE, a novel ego-centric framework for graph homomorphic encryption that addresses the limitations of existing systems by enabling efficient inference on large-scale financial graphs. By leveraging a template phenomenon to canonicalize ego-graphs and packing structurally identical trees into shared CKKS ciphertexts, TGHE allows for SIMD-parallel encrypted inference. The method achieves a remarkable 66.9x speedup on a dataset with 3.7 million nodes while maintaining minimal accuracy loss, demonstrating its effectiveness in dynamic environments.
Achieving a 66.9x speedup in encrypted GNN inference opens the door to real-time privacy-preserving analytics on massive financial graphs.
Existing homomorphic encryption (HE)-based GNN systems adopt a graph-centric paradigm that couples per-query cost to global graph size, limiting evaluations to at most ~20k nodes and making them incompatible with dynamic, large-scale financial graphs. We propose TGHE (Template-based Graph Homomorphic Encryption), an ego-centric framework that resolves this by exploiting a template phenomenon: local computation trees in transaction graphs converge into a small set of structural shapes. TGHE canonicalizes ego-graphs at the edge and packs structurally identical trees into shared CKKS ciphertexts for SIMD-parallel encrypted inference, with two long-tail optimizers (Approximate Template Fitting and Topology Collapse) ensuring full SIMD coverage. On DGraphFin (3.7M nodes, 4.3M edges), TGHE-Collapse achieves a 66.9x speedup over the sequential encrypted baseline with less than 0.002 AUC loss.
