Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2510.26709 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914133269544960 |
|---|---|
| author | Chen, Chuyan Ma, Chenyang Li, Zhangxin He, Yutong Dong, Yanjie Yuan, Kun |
| author_facet | Chen, Chuyan Ma, Chenyang Li, Zhangxin He, Yutong Dong, Yanjie Yuan, Kun |
| contents | Communication remains a central bottleneck in large-scale distributed machine learning, and gradient sparsification has emerged as a promising strategy to alleviate this challenge. However, existing gradient compressors face notable limitations: Rand-$K$ discards structural information and performs poorly in practice, while Top-$K$ preserves informative entries but loses the contraction property and requires costly All-Gather operations. In this paper, we propose ARC-Top-$K$, an {All-Reduce}-Compatible Top-$K$ compressor that aligns sparsity patterns across nodes using a lightweight sketch of the gradient, enabling index-free All-Reduce while preserving globally significant information. ARC-Top-$K$ is provably contractive and, when combined with momentum error feedback (EF21M), achieves linear speedup and sharper convergence rates than the original EF21M under standard assumptions. Empirically, ARC-Top-$K$ matches the accuracy of Top-$K$ while reducing wall-clock training time by up to 60.7\%, offering an efficient and scalable solution that combines the robustness of Rand-$K$ with the strong performance of Top-$K$. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_26709 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | An All-Reduce Compatible Top-K Compressor for Communication-Efficient Distributed Learning Chen, Chuyan Ma, Chenyang Li, Zhangxin He, Yutong Dong, Yanjie Yuan, Kun Machine Learning Distributed, Parallel, and Cluster Computing Communication remains a central bottleneck in large-scale distributed machine learning, and gradient sparsification has emerged as a promising strategy to alleviate this challenge. However, existing gradient compressors face notable limitations: Rand-$K$ discards structural information and performs poorly in practice, while Top-$K$ preserves informative entries but loses the contraction property and requires costly All-Gather operations. In this paper, we propose ARC-Top-$K$, an {All-Reduce}-Compatible Top-$K$ compressor that aligns sparsity patterns across nodes using a lightweight sketch of the gradient, enabling index-free All-Reduce while preserving globally significant information. ARC-Top-$K$ is provably contractive and, when combined with momentum error feedback (EF21M), achieves linear speedup and sharper convergence rates than the original EF21M under standard assumptions. Empirically, ARC-Top-$K$ matches the accuracy of Top-$K$ while reducing wall-clock training time by up to 60.7\%, offering an efficient and scalable solution that combines the robustness of Rand-$K$ with the strong performance of Top-$K$. |
| title | An All-Reduce Compatible Top-K Compressor for Communication-Efficient Distributed Learning |
| topic | Machine Learning Distributed, Parallel, and Cluster Computing |
| url | https://arxiv.org/abs/2510.26709 |