Research map / Optimization / Distributed and compressed
Preserved central model for faster bidirectional compression in distributed settings
Constantin Philippenko, Aymeric Dieuleveut
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 |
|---|---|---|---|
| add downlink compression on top of uplink compression (MCM vs Diana) | Diana · w8a | log10 excess loss after 450 epochs: -3.0 -> -2.5 | Hurt |
| add downlink compression (MCM vs Diana) | Diana · a9a | log10 excess loss after 450 epochs: -2.7 -> -2.7 | No clear effect |
| add downlink compression (MCM vs Diana) | Diana · quantum | log10 excess loss after 450 epochs: -3.2 -> -3.2 | No clear effect |
| add downlink compression (MCM vs Diana) | Diana · phishing | log10 excess loss after 450 epochs: -3.5 -> -3.4 | No clear effect |
| MCM instead of Dore for bidirectional compression | Dore · a9a | log10 excess loss after 450 epochs: -1.8 -> -2.7 | Improved |
| MCM instead of Dore for bidirectional compression | Dore · w8a | log10 excess loss after 450 epochs: -1.75 -> -2.5 | Improved |
| MCM instead of Artemis for bidirectional compression | Artemis · CIFAR-10 | accuracy after 300 epochs: 54.8 -> 63.5 | Improved |
| MCM instead of Dore for bidirectional compression | Dore · Fashion MNIST | accuracy after 300 epochs: 87.9 -> 90.6 | Improved |
| MCM versus Diana in deep learning | Diana · Fashion MNIST | accuracy after 300 epochs: 92.4 -> 90.6 | Hurt |
| MCM versus Diana in deep learning | Diana · Fashion MNIST | train loss after 300 epochs: 0.141 -> 0.209 | Hurt |
| MCM versus Diana in deep learning | Diana · MNIST | accuracy after 300 epochs: 98.9 -> 98.8 | No clear effect |
| add randomization of the model (Rand-MCM) (ablation) | MCM · least-squares and logistic regression | convergence stability | Improved |
Closest papers by meaning
- Lower Bounds and Nearly Optimal Algorithms in Distributed Learning with Communication Compression (NeurIPS 2022 · similarity 0.91)
- 2Direction: Theoretically Faster Distributed Training with Bidirectional Communication Compression (NeurIPS 2023 · similarity 0.89)
- Scalable Distributed Stochastic Optimization via Bidirectional Compression: Beyond Pessimistic Limits (NeurIPS 2026 · similarity 0.88)
- Improving the Worst-Case Bidirectional Communication Complexity for Nonconvex Distributed Optimization under Function Similarity (NeurIPS 2024 · similarity 0.88)
- Accelerated Methods with Compressed Communications for Distributed Optimization Problems Under Data Similarity (AAAI 2025 · similarity 0.88)
- Towards Faster Decentralized Stochastic Optimization with Communication Compression (ICLR 2025 · similarity 0.87)
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.
