Research map / Architectures and efficiency / RNNs and normalization
Consensus Control for Decentralized 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 |
|---|---|---|---|
| Reduce phase-1 consensus distance from Xi max to 1/4 Xi max (ablation) | ResNet-20 · CIFAR-10, n=32, ring | test top-1 accuracy: 91.78 ± 0.35 -> 92.74 ± 0.10 | Helped |
| Reduce phase-1 consensus distance from Xi max to 1/2 Xi max (ablation) | ResNet-20 · CIFAR-10, n=32, ring | test top-1 accuracy: 91.78 ± 0.35 -> 92.36 ± 0.21 | Helped |
| Reduce phase-1 consensus distance below 1/4 Xi max (ablation) | ResNet-20 · CIFAR-10, n=32, ring | test top-1 accuracy: 92.74 for 1/4 Xi max; 92.77 for 1/8 Xi max; 92.72 for 1/16 Xi max | No clear effect |
| Allow non-negligible consensus distance in phase 2 rather than using centralized averaging (ablation) | ResNet-20 · CIFAR-10, n=32, ring | test top-1 accuracy: 92.82 -> 93.04 | Helped |
| Reduce phase-2 consensus distance from Xi max to 1/4 Xi max (ablation) | ResNet-20 · CIFAR-10, n=32, ring | test top-1 accuracy: 93.04 ± 0.01 -> 92.87 ± 0.11 | Hurt |
| Reduce phase-3 consensus distance from Xi max to 1/4 Xi max (ablation) | ResNet-20 · CIFAR-10, n=32, ring | test top-1 accuracy: 92.60 ± 0.00 -> 92.85 ± 0.24 | Helped |
| Reduce phase-1 consensus distance from Xi max to 1/2 Xi max (ablation) | ResNet-20-3 · ImageNet-32, n=16, ring | test top-1 accuracy: 51.22 ± 0.08 -> 51.79 ± 0.10 | Helped |
| Reduce phase-1 consensus distance from Xi max to 1/4 Xi max (ablation) | ResNet-20-3 · ImageNet-32, n=16, ring | test top-1 accuracy: 51.22 ± 0.08 -> 51.71 ± 0.03 | Helped |
| Change consensus distance in phase 3 (ablation) | ResNet-20-3 · ImageNet-32, n=16, ring | test top-1 accuracy: 51.80 ± 0.10; 51.81 ± 0.13; 51.81 ± 0.04 | No clear effect |
| Use longer phase-1 training instead of reducing phase-1 consensus distance (ablation) | ResNet-20 · CIFAR-10, n=32, ring | test top-1 accuracy: 91.78 ± 0.35 -> 92.04 ± 0.14; 92.36 ± 0.21 -> 92.67 ± 0.13 | Mixed |
| Control only the initial phase and leave later phases decentralized (ablation) | ResNet-20 · CIFAR-10, n=32, ring | test top-1 accuracy: 91.78 ± 0.35 -> 92.81 ± 0.22 | Helped |
| Use local-estimate consensus control with q=1e{-}4 for phase 1 (ablation) | ResNet-20 · CIFAR-10, n=32, ring | test top-1 accuracy: 91.78 ± 0.35 -> 92.85 ± 0.16 | Helped |
Closest papers by meaning
- From Promise to Practice: Realizing High-performance Decentralized Training (ICLR 2025 · similarity 0.92)
- On The Surprising Effectiveness of a Single Global Merging in Decentralized Learning (ICLR 2026 · similarity 0.88)
- A Decentralized Parallel Algorithm for Training Generative Adversarial Nets (NeurIPS 2020 · similarity 0.88)
- DIMAT: Decentralized Iterative Merging-And-Training for Deep Learning Models (CVPR 2024 · similarity 0.88)
- Unveiling the Power of Multiple Gossip Steps: A Stability-Based Generalization Analysis in Decentralized Training (NeurIPS 2025 · similarity 0.87)
- Decentralized Diffusion Models (CVPR 2025 · similarity 0.86)
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.