Back to feed
arXiv cs.LG·

Worker Disagreement Reveals Sharp Directions in Local SGD

Signal
75
Hype
15
In three linesResearchers show Local SGD exposes anisotropic loss geometry through worker disagreement. Worker-average gaps provide a Hessian-free estimator of dominant spectral directions. Validated on MLPs, CNNs, and Transformers.
Read source
Your take?
PapersReinforcement learning

Summary generated by Claude — human-verified