Federated Sinkhorn: Communication, Performance, and Privacy for Distributed Entropic Optimal Transport
Jeremy Kulcsar, Vyacheslav Kungurtsev, Georgios Korpas, Giulio Giaconi, William Shoosmith · 2025
Read Federated Sinkhorn: Communication, Performance, and Privacy for Distributed Entropic Optimal Transport by Jeremy Kulcsar, Vyacheslav Kungurtsev, Georgios Korpas, Giulio Giaconi, William Shoosmith (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-07021.txt.
Description
Jeremy Kulcsar and co-authors wrote this 2025 paper on running a classic transport algorithm across machines that cannot pool their data. Sinkhorn iterations solve entropy-regularized optimal transport, but the kernel matrix here is split row-wise among workers. The team design two exact protocols that pass only slices of the scaling vectors around.
One protocol has every worker talk to every other; the other routes through a central server. For both, the authors derive closed-form costs for computation, communication, and memory, and prove the distributed iterates match the centralized algorithm exactly. Experiments on CPU and GPU clusters confirm the cost model, and show the repeated global exchange becoming the bottleneck as workers multiply. Optional extras cover a bounded-delay asynchronous mode and a privacy layer.
Catalog record
| CREATOR | Jeremy Kulcsar, Vyacheslav Kungurtsev, Georgios Korpas, Giulio Giaconi, William Shoosmith |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 95 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-07021