Research map / Federated learning and privacy / Privacy attacks and secure learning
Glyph: Fast and Accurately Training Deep Neural Networks on Encrypted Data
Qian Lou, Bo Feng, Geoffrey Charles Fox, 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 BGV lookup-table sigmoid activations with TFHE ReLU/softmax and switch between BGV and TFHE (ablation) | FHESGD-based MLP · Encrypted MNIST | mini-batch training latency: reduces latency by 97.4% | Improved |
| Use BGV instead of BFV for linear layers in the switching scheme (ablation) | Glyph MLP · Encrypted MNIST | mini-batch training latency: 76.4% decrease vs Chimera | Improved |
| Use CNN with transfer learning (fixed SVHN-pretrained conv layers) instead of MLP (ablation) | Glyph · Encrypted MNIST | training latency: 56.7% decrease | Improved |
| Use CNN with transfer learning instead of MLP (ablation) | Glyph · Encrypted MNIST | test accuracy: ~2% | Improved |
| Add transfer learning (SVHN features) to Glyph CNN (ablation) | Glyph CNN · MNIST after 5 epochs | test accuracy (%): 97.1 -> 98.6 | Improved |
| Add transfer learning (CIFAR-10 features) to Glyph CNN (ablation) | Glyph CNN · Skin-Cancer-MNIST | test accuracy: +4% test accuracy boost | Improved |
| Use CNN architecture instead of FHESGD MLP without transfer learning (ablation) | Glyph CNN · MNIST after 5 epochs | test accuracy (%): 96.4 -> 97.1 | Improved |
| Use BGV-TFHE Glyph CNN instead of BFV-TFHE Chimera CNN | Glyph · Encrypted MNIST | training latency: 69% reduction; 28.6 days -> 8 days | Improved |
Closest papers by meaning
- HEPrune: Fast Private Training of Deep Neural Networks With Encrypted Data Pruning (NeurIPS 2024 · similarity 0.84)
- ReBoot: Encrypted Training of Deep Neural Networks with CKKS Bootstrapping (AAAI 2026 · similarity 0.83)
- EncryptedLLM: Privacy-Preserving Large Language Model Inference via GPU-Accelerated Fully Homomorphic Encryption (ICML 2025 · similarity 0.81)
- DCT-CryptoNets: Scaling Private Inference in the Frequency Domain (ICLR 2025 · similarity 0.80)
- FicGCN: Unveiling the Homomorphic Encryption Efficiency from Irregular Graph Convolutional Networks (ICML 2025 · similarity 0.80)
- Rethinking DP-SGD in Discrete Domain: Exploring Logistic Distribution in the Realm of signSGD (ICML 2024 · similarity 0.80)
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.
