Research map / Optimization / Distributed and compressed
Beyond Euclidean Gossip: KL-Barycentric Consensus on Heterogeneous and Imbalanced Images
Lu Xu, Guosheng Yin
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 |
|---|---|---|---|
| replace Euclidean parameter mixing with KL-barycentric consensus in expectation space (ablation) | KL-consensus Adam · CIFAR-100, 8 clients, Dirichlet 0.01 | test accuracy (%): 35.18 -> 54.39 | Improved |
| replace Euclidean mixing with KL-barycentric consensus (ablation) | KL-consensus Adam · CIFAR-100, 8 clients, Dirichlet 1.0 | test accuracy (%): 53.72 -> 60.10 | Improved |
| replace Euclidean mixing with KL-barycentric consensus under size imbalance 0.02 (ablation) | KL-consensus Adam · CIFAR-100, 8 clients, imbalance 0.02 | test accuracy (%): 44.87 -> 52.82 | Improved |
| use KL-consensus Adam instead of QGM | KL-consensus Adam · CIFAR-100, 8 clients, Dirichlet 0.01 | test accuracy (%): 40.85 -> 54.39 | Improved |
| use KL-consensus Adam instead of QGM with 16 clients at mild heterogeneity | KL-consensus Adam · CIFAR-100, 16 clients, Dirichlet 1.0 | test accuracy (%): 43.02 -> 43.68 | No clear effect |
| use KL-consensus Adam instead of QGM | KL-consensus Adam · Kvasir-SEG, Dirichlet 0.1 | Dice: 0.720 -> 0.767 | Improved |
| add noisy (JPEG-compressed) clients | KL-consensus Adam · Kvasir-SEG, Dirichlet 0.1 | Dice: 0.767 -> 0.758 (1 noisy), 0.750 (3 noisy) | Hurt |
| change communication graph from ring to 2-rings or tree (ablation) | KL-consensus Adam · Kvasir-SEG | Dice: 0.766 -> 0.769 (2-rings), 0.759 (tree) | No clear effect |
Closest papers by meaning
- $\textbf{A}^2\textbf{CiD}^2$: Accelerating Asynchronous Communication in Decentralized Deep Learning (NeurIPS 2023 · similarity 0.86)
- DPA: Decentralized Primal Averaging with Quasi-Global Momentum for Highly Heterogeneous Data (NeurIPS 2026 · similarity 0.86)
- DIMAT: Decentralized Iterative Merging-And-Training for Deep Learning Models (CVPR 2024 · similarity 0.85)
- Robust Distributed Gradient Aggregation Using Projections onto Gradient Manifolds (AAAI 2024 · similarity 0.84)
- From Promise to Practice: Realizing High-performance Decentralized Training (ICLR 2025 · similarity 0.83)
- A Quadratic Synchronization Rule for Distributed Deep Learning (ICLR 2024 · 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.
