Research map / Architectures and efficiency / Pruning and efficient tuning
SpENCNN: Orchestrating Encoding and Sparsity for Fast Homomorphically Encrypted Neural Network Inference
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| HE-group convolution with group-interleaved encoding (adding more groups up to Gbase) (ablation) | LeNet-like, VGG-5, HEFNet, ResNet-20 · CKKS HE inference latency | Speedup (×): LeNet 1.86x (2 groups) to 4.16x (8 groups); HEFNet up to 10.21x (16 groups) | Helped |
| HE-group convolution alone, pushed beyond Gbase (ablation) | HEFNet · CIFAR-10 HE inference | Accuracy (%): 83.67% at 8 groups to 80.06% at 16 groups | Hurt |
| HE-group convolution applied to the VGG-5 FC-heavy model (ablation) | VGG-5 · CIFAR-10 HE inference | Speedup (×): limited to 1.19x at 8 groups | Hurt |
| Sub-block pruning (HE-friendly structured sub-block removal) alone (ablation) | LeNet-like, VGG-5, HEFNet, ResNet-20 · CKKS HE inference latency | Speedup (×): 2.62x, 6.15x, 2.57x, 1.36x | Helped |
| Non-structured pruning instead of sub-block pruning (ablation) | LeNet-like, VGG-5, HEFNet, ResNet-20 · CKKS HE inference latency | Speedup (×): ~1.03x to 1.14x | No clear effect |
| Structured channel pruning instead of sub-block pruning (ablation) | LeNet-like, VGG-5, HEFNet, ResNet-20 · CKKS HE inference latency | Speedup (×): ~1.03x to 1.13x | No clear effect |
| Combined HE-group convolution + sub-block pruning vs. Hunter structured pruning | LeNet-like, VGG-5, HEFNet, ResNet-20 · CKKS HE inference speedup at matched accuracy | Speedup (×): e.g., HEFNet 19.26x vs 2.22x; ResNet-20 1.87x vs 1.40x | Helped |
Closest papers by meaning
- FicGCN: Unveiling the Homomorphic Encryption Efficiency from Irregular Graph Convolutional Networks (ICML 2025 · similarity 0.88)
- PrivCirNet: Efficient Private Inference via Block Circulant Transformation (NeurIPS 2024 · similarity 0.88)
- HEPrune: Fast Private Training of Deep Neural Networks With Encrypted Data Pruning (NeurIPS 2024 · similarity 0.88)
- Converting Transformers to Polynomial Form for Secure Inference Over Homomorphic Encryption (ICML 2024 · similarity 0.87)
- DCT-CryptoNets: Scaling Private Inference in the Frequency Domain (ICLR 2025 · similarity 0.87)
- CipherPrune: Efficient and Scalable Private Transformer Inference (ICLR 2025 · similarity 0.87)
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.