Research map / Optimization / Distributed and compressed
Epidemic Learning: Boosting Decentralized Learning with Randomized Communication
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 static 7-regular topology with randomly changing topology each round (EL-Oracle) | D-PSGD · CIFAR-10 non-IID | top-1 test accuracy (%): 64.32 -> 66.56 | Helped |
| replace static random topology with randomized dynamic topology | D-PSGD · CIFAR-10 non-IID, 3000 rounds | top-1 test accuracy (%): 63.4 -> 65.3 | Helped |
| locally sampled targets (EL-Local) instead of coordinated s-regular graph (EL-Oracle) (ablation) | Epidemic Learning · CIFAR-10 non-IID | top-1 test accuracy (%): 66.56 -> 66.14 | No clear effect |
| EL-Oracle vs 7-Regular static communication to reach target accuracy | D-PSGD · CIFAR-10 non-IID | communication per node (GiB): 7.03 -> 4.12 | Helped |
| increase sample size s from 4 to 7 to 14 (ablation) | EL-Oracle and EL-Local · CIFAR-10 non-IID | convergence speed (accuracy vs rounds): communication 4.3 GiB (s=4) -> 15.1 GiB (s=14) | Helped |
| use 8-U-EquiStatic topology instead of 7-Regular static | D-PSGD · CIFAR-10 non-IID | top-1 test accuracy (%): 64.32 -> 62.72 | Hurt |
| static unlucky topology (two partitions linked by two edges) vs static random 7-regular | D-PSGD · CIFAR-10 non-IID, 3000 rounds | top-1 test accuracy (%): 63.4 -> 61.8 | Hurt |
Closest papers by meaning
- Robust and Scalable Collaborative Learning via Pull-Based Epidemic Communication (NeurIPS 2026 · similarity 0.90)
- Unveiling the Power of Multiple Gossip Steps: A Stability-Based Generalization Analysis in Decentralized Training (NeurIPS 2025 · similarity 0.86)
- RESIST: Resilient Decentralized Learning Using Consensus Gradient Descent (NeurIPS 2026 · similarity 0.86)
- Decentralized Sporadic Federated Learning: A Unified Algorithmic Framework with Convergence Guarantees (ICLR 2025 · similarity 0.86)
- PDUDT: Provable Decentralized Unlearning under Dynamic Topologies (ICML 2025 · similarity 0.85)
- Revisiting Differentially Private Algorithms for Decentralized Online Learning (ICML 2025 · similarity 0.85)
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.