Research map / Architectures and efficiency / RNNs and normalization
LinGCN: Structural Linearized Graph Convolutional Network for Homomorphically Encrypted Inference
Hongwu Peng, Ran Ran, Yukui Luo, Jiahui Zhao, Shaoyi Huang, Kiran Thorat, Tong Geng, Chenghong Wang, Xiaolin Xu, Wujie Wen, Caiwen Ding
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use LinGCN instead of CryptoGCN at 6 nonlinear layers | STGCN-3-128 · NTU-XView | test accuracy (%): 74.25 -> 77.55 | Improved |
| use LinGCN instead of CryptoGCN at 6 nonlinear layers | STGCN-3-128 · NTU-XView | latency (s): 4273.89 -> 1856.95 | Improved |
| use LinGCN instead of CryptoGCN at 6 nonlinear layers | STGCN-3-256 · NTU-XView | test accuracy (%): 75.31 -> 80.29 | Improved |
| use LinGCN instead of CryptoGCN at 6 nonlinear layers | STGCN-3-256 · NTU-XView | latency (s): 10580.41 -> 4632.05 | Improved |
| reduce the effective nonlinear layers from 6 to 2 | LinGCN STGCN-3-128 · NTU-XView | latency (s): 1856.95 -> 741.55 | Improved |
| reduce the effective nonlinear layers from 6 to 2 | LinGCN STGCN-3-128 · NTU-XView | test accuracy (%): 77.55 -> 75.16 | Hurt |
| reduce the effective nonlinear layers from 6 to 4 | LinGCN STGCN-6-256 · NTU-XView | test accuracy (%): 84.52 -> 85.78 | Improved |
| reduce nonlinear layers from 6 to 2 | LinGCN · Flickr | latency (s): 4290.93 -> 2740.94 | Improved |
| reduce nonlinear layers from 6 to 2 | LinGCN · Flickr | accuracy (val/test, %): 0.5281/0.5275 -> 0.5247/0.5266 | Hurt |
| perform nonlinear reduction before polynomial replacement (ablation) | LinGCN STGCN-3-256 · NTU-XView | accuracy (%): more than 2% | Improved |
| use structural rather than layer-wise linearization (ablation) | LinGCN STGCN-3-256 · NTU-XView, 2 to 5 effective nonlinear layers | accuracy (%): 1.5% | Improved |
| set distillation KL weight η to 0.2 (ablation) | LinGCN STGCN-3-256, 6 nonlinear layers · NTU-XView | accuracy (%) | Improved |
Closest papers by meaning
- FicGCN: Unveiling the Homomorphic Encryption Efficiency from Irregular Graph Convolutional Networks (ICML 2025 · similarity 0.91)
- GRAIN: Exact Graph Reconstruction from Gradients (ICLR 2025 · similarity 0.86)
- DCT-CryptoNets: Scaling Private Inference in the Frequency Domain (ICLR 2025 · similarity 0.86)
- PrivCirNet: Efficient Private Inference via Block Circulant Transformation (NeurIPS 2024 · similarity 0.85)
- Converting Transformers to Polynomial Form for Secure Inference Over Homomorphic Encryption (ICML 2024 · similarity 0.85)
- CipherPrune: Efficient and Scalable Private Transformer Inference (ICLR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
