Research map / Architectures and efficiency / RNNs and normalization
SHE: A Fast and Accurate Deep Neural Network for Encrypted Data
Qian Lou, Lei Jiang
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 |
|---|---|---|---|
| Replace polynomial activations and pooling with TFHE ReLU and max pooling (ablation) | FCN-style network · MNIST | accuracy: 98.71 -> 99.54 | Improved |
| Replace polynomial activations and pooling with TFHE ReLU and max pooling (ablation) | FCN-style network · MNIST | leveled inference latency: 39.1 -> 88.6 | Hurt |
| Add logarithmic quantization and replace matrix multiplications with shift-accumulation (ablation) | TCN · MNIST | leveled inference latency: 88.6 -> 9.3 | Improved |
| Add logarithmic quantization and replace matrix multiplications with shift-accumulation (ablation) | TCN · CIFAR-10 | leveled inference latency: 31K -> 2258 | Improved |
| Use a deeper network architecture (ablation) | SHE · MNIST | accuracy: 99.54 -> 99.77 | Improved |
| Use a deeper ResNet-18 architecture (ablation) | SHE · CIFAR-10 | accuracy: 92.54 -> 94.62 | Improved |
| Increase the polynomial approximation activation degree | CNT · MNIST | accuracy: 98.95 -> 99.52 | Improved |
| Increase the polynomial approximation activation degree | CNT · MNIST | inference latency: 250 -> 320 | Hurt |
| Quantize weights, inputs, and activations to 1 bit | MNIST CNN · MNIST | accuracy: 98.95 -> 93.71 | Hurt |
| Replace polynomial activations and pooling with TFHE ReLU and max pooling (ablation) | FCN-style network · CIFAR-10 | accuracy: 76.72 -> 92.54 | Improved |
| Add logarithmic quantization and replace matrix multiplications with shift-accumulation (ablation) | TCN · CIFAR-10 | accuracy: 92.54 -> 92.54 | No clear effect |
| Use SHE's shift-accumulation operations in place of matrix multiplications (ablation) | TFHE CNN · ImageNet AlexNet | multiplicative depth: 42.3K -> 6.8K | Improved |
Closest papers by meaning
- Converting Transformers to Polynomial Form for Secure Inference Over Homomorphic Encryption (ICML 2024 · similarity 0.86)
- FicGCN: Unveiling the Homomorphic Encryption Efficiency from Irregular Graph Convolutional Networks (ICML 2025 · similarity 0.84)
- Encryption-Friendly LLM Architecture (ICLR 2025 · similarity 0.84)
- DCT-CryptoNets: Scaling Private Inference in the Frequency Domain (ICLR 2025 · similarity 0.83)
- PrivCirNet: Efficient Private Inference via Block Circulant Transformation (NeurIPS 2024 · similarity 0.83)
- HEPrune: Fast Private Training of Deep Neural Networks With Encrypted Data Pruning (NeurIPS 2024 · similarity 0.83)
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.
