Beyond Scaffold: A Unified Spatio-Temporal Gradient Tracking Method footnoteinfo The computations/data handling/[SIMILAR] were/was enabled by resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS), partially funded by the Swedish Research Council through gran
Yan Huang, Jinming Xu, Jiming Chen, Karl Henrik Johansson · 2025
Read Beyond Scaffold: A Unified Spatio-Temporal Gradient Tracking Method footnoteinfo The computations/data handling/[SIMILAR] were/was enabled by resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS), partially funded by the Swedish Research Council through gran by Yan Huang, Jinming Xu, Jiming Chen, Karl Henrik Johansson (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-2512-01732.txt.
Description
Yan Huang and co-authors track gradients in distributed learning beyond scaffold methods. Local updates between communication rounds cut network cost. Data gaps across nodes pull local models apart, though. Local gradient noise adds more drift. Both effects move models from the global best.
Their unified spatio-temporal method tracks gradients across space and time. Tracking corrects drift while local steps continue. One frame covers both distributed and federated settings. Less drift means fewer rounds to reach a target, which saves bandwidth for edge and phone fleets. Spatio-temporal tracking reuses stale gradients with care, turning a known source of error into a signal that speeds convergence.
Catalog record
| CREATOR | Yan Huang, Jinming Xu, Jiming Chen, Karl Henrik Johansson |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 63 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-01732