Research map / Federated learning and privacy / Communication-efficient FL
CoPriv: Network/Protocol Co-Optimization for Communication-Efficient Private Inference
Wenxuan Zeng, Meng Li, Haichuan Yang, Wen-jie Lu, Runsheng Wang, Ru Huang
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 |
|---|---|---|---|
| rewrite Winograd EWMM as GEMM with tile aggregation (ablation) | Winograd conv protocol · ResNet-18 block | communication (GB): 256 -> 10.45 | Improved |
| add DNN-aware adaptive sender selection (ablation) | Winograd conv protocol · ResNet-18 block | communication (GB): 10.45 -> 7.76 | Improved |
| naive EWMM Winograd instead of regular conv (ablation) | CrypTFlow2 conv · ResNet-18 block | communication (GB): 23.74 -> 256 | Hurt |
| add Winograd protocol to SNL ResNet-18 (ablation) | SNL · CIFAR-100 | total communication (GB): 61.45 -> 33.84 | Improved |
| add Winograd protocol to SNL MobileNetV2 (ablation) | SNL · CIFAR-100 | total communication (GB): 7.710 -> 7.132 | Improved |
| add network re-parameterization after pruning (ablation) | CoPriv MobileNetV2 · CIFAR-100 | total communication (GB): 7.806 -> 4.197 | Improved |
| add re-parameterization after pruning (ablation) | CoPriv MobileNetV2 · CIFAR-100 | online communication (GB): 0.606 -> 0.331 | Improved |
| add Winograd protocol after re-parameterization (ablation) | CoPriv MobileNetV2 · CIFAR-100 | total communication (GB): 4.197 -> 3.154 | Improved |
| add Winograd protocol after re-parameterization (ablation) | CoPriv MobileNetV2 · CIFAR-100 | online communication (GB): 0.331 -> 0.331 | No clear effect |
| use communication-aware regularization L_comm in pruning (ablation) | CoPriv pruning · CIFAR-100 per-block ReLU budget | pruning distribution | Improved |
| CoPriv vs SNL on MobileNetV2-w1.4 | CoPriv · ImageNet | accuracy and communication: 1.4% higher accuracy, 9.98x online, 3.88x total | Improved |
| CoPriv vs SNL on ResNet-18 | CoPriv · CIFAR-100 | accuracy and communication: 4.3% higher accuracy, 1.55x online, 19.5x total | Improved |
Closest papers by meaning
- United We Stand: Accelerating Privacy-Preserving Neural Inference by Conjunctive Optimization with Interleaved Nexus (AAAI 2024 · similarity 0.86)
- Jaguar: Fast Private CNN Inference with Power-of-Two Homomorphic Arithmetic (NeurIPS 2026 · similarity 0.86)
- PrivCirNet: Efficient Private Inference via Block Circulant Transformation (NeurIPS 2024 · similarity 0.85)
- DCT-CryptoNets: Scaling Private Inference in the Frequency Domain (ICLR 2025 · similarity 0.84)
- CipherPrune: Efficient and Scalable Private Transformer Inference (ICLR 2025 · similarity 0.84)
- PrivDNFIS: Privacy-preserving and Efficient Deep Neuro-Fuzzy Inference System (AAAI 2025 · 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.
