Research map / Federated learning and privacy / Communication-efficient FL
Global Update Tracking: A Decentralized Learning Algorithm for Heterogeneous Data
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 |
|---|---|---|---|
| replace DSGD's update with Global Update Tracking | Decentralized training · CIFAR-10, alpha=0.1, 16-agent ring, ResNet-20 | test accuracy: 72.21± 2.37 -> 81.86 ± 1.99 | Helped |
| replace QG-DSGDm with QG-GUTm | Decentralized training · CIFAR-10, alpha=0.1, 16-agent ring, ResNet-20 | test accuracy: 84.21± 2.12 -> 86.44 ± 0.36 | Helped |
| replace QG-DSGDm with QG-GUTm | Decentralized training · CIFAR-10, alpha=1, 16-agent ring, ResNet-20 | test accuracy: 88.23 ± 0.51 -> 88.22 ± 0.36 | No clear effect |
| replace QG-DSGDm with QG-GUTm | Decentralized training · CIFAR-10, alpha=0.01, 32-agent ring, ResNet-20 | test accuracy: 69.99 ± 0.60 -> 72.04 ± 3.18 | Helped |
| replace QG-DSGDm with QG-GUTm | Decentralized training · CIFAR-100, alpha=0.01, 16-agent ring, ResNet-20 | test accuracy: 44.17 ± 3.64 -> 50.45 ± 1.30 | Helped |
| replace QG-DSGDm with QG-GUTm | Decentralized training · Fashion MNIST, alpha=0.1, 16-agent ring, LeNet-5 | test accuracy: 89.94 ± 0.44 -> 90.11 ± 0.02 | No clear effect |
| track model updates without changing the received neighbor tracking variable's reference (ablation) | GUT · CIFAR-10, alpha=0.1, 16-agent ring, ResNet-20 | test accuracy: 81.86 ± 1.99 -> 72.78 ± 0.80 | Hurt |
| use neighborhood-update difference as a bias correction (ablation) | GUT · CIFAR-10, alpha=0.1, 16-agent ring, ResNet-20 | test accuracy: 81.86 ± 1.99 -> 72.62 ± 1.16 | Hurt |
| add GUT-style tracking to the decentralized update | Gradient Tracking · CIFAR-10, alpha=0.1, 16-agent ring, ResNet-20 | test accuracy: 80.61± 2.41 -> 81.86 ± 1.99 | Helped |
| add local momentum to GUT (ablation) | GUT · CIFAR-10, alpha=0.1, 16-agent ring, ResNet-20 | test accuracy: 81.86± 1.99 -> 79.95± 1.67 | Hurt |
| add quasi-global momentum to GUT (ablation) | GUT · CIFAR-10, alpha=0.1, 16-agent ring, ResNet-20 | test accuracy: 81.86± 1.99 -> 86.44 ± 0.36 | Helped |
| add Nesterov momentum to QG-GUTm (ablation) | QG-GUTm · CIFAR-10, alpha=0.1, 16-agent ring, ResNet-20 | test accuracy: 86.44 ± 0.36 -> 86.55 ± 0.49 | No clear effect |
Closest papers by meaning
- DPA: Decentralized Primal Averaging with Quasi-Global Momentum for Highly Heterogeneous Data (NeurIPS 2026 · similarity 0.87)
- Quasi-global Momentum: Accelerating Decentralized Deep Learning on Heterogeneous Data (ICML 2021 · similarity 0.87)
- Heterogeneous Decentralized Diffusion Models (CVPR 2026 · similarity 0.86)
- DIMAT: Decentralized Iterative Merging-And-Training for Deep Learning Models (CVPR 2024 · similarity 0.85)
- From Promise to Practice: Realizing High-performance Decentralized Training (ICLR 2025 · similarity 0.85)
- Decentralized Diffusion Models (CVPR 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.