Research map / Architectures and efficiency / CNN design
Secure Quantized Training for Deep Learning
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 |
|---|---|---|---|
| Use AMSGrad instead of SGD (ablation) | MNIST Network C (LeNet), three-party LAN · MNIST | accuracy: 98.5% -> 99.0% | Helped |
| Use AMSGrad instead of SGD (ablation) | MNIST Networks A, B, and D, three-party LAN · MNIST | accuracy: 97.8% -> 98.1%; 98.0% -> 98.6%; 98.1% -> 98.5% | Helped |
| Use AMSGrad instead of SGD (ablation) | CIFAR-10 AlexNet-based network · CIFAR-10 | accuracy: 64.9% -> 63.6% | Hurt |
| Use Adam instead of SGD (ablation) | CIFAR-10 AlexNet-based network · CIFAR-10 | accuracy: 64.9% -> 64.7% | Hurt |
| Use 16-bit precision with nearest rounding instead of probabilistic rounding (ablation) | MNIST training with SGD · MNIST | accuracy | Hurt |
| Increase precision beyond f=16 (ablation) | MNIST training · MNIST | accuracy | No clear effect |
| Reduce precision to f=8 (ablation) | MNIST training · MNIST | training convergence | Hurt |
| Add a Dropout layer (ablation) | MNIST training · MNIST | accuracy | Helped |
| Replace nearest rounding with probabilistic rounding (ablation) | Quantized matrix multiplication in secure training · MNIST | accuracy | Helped |
| Use the proposed exponentiation protocol instead of the prior algorithm (ablation) | MPC exponentiation · Communication across security models | communication: 27 -> 16; 498 -> 323; 1,338 -> 813; 214,476 -> 121,747 | Helped |
| Optimize inverse square root with the proposed XOR-based least-significant-bit computation (ablation) | MPC inverse square root · Communication across security models | communication: 19 -> 9; 160 -> 114; 481 -> 342; 25,456 -> 21,522 | Helped |
Closest papers by meaning
- Ditto: Quantization-aware Secure Inference of Transformers upon MPC (ICML 2024 · similarity 0.86)
- CrypTen: Secure Multi-Party Computation Meets Machine Learning (NeurIPS 2021 · similarity 0.86)
- Training Quantized Nets: A Deeper Understanding (NeurIPS 2017 · similarity 0.85)
- DPQuant: Efficient and Private Model Training via Dynamic Quantization Scheduling (ICLR 2026 · similarity 0.84)
- Towards Efficient Verification of Quantized Neural Networks (AAAI 2024 · similarity 0.84)
- ReBoot: Encrypted Training of Deep Neural Networks with CKKS Bootstrapping (AAAI 2026 · similarity 0.84)
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.