World Wires · story 28601 · corroborated · 1 source(s)

Decentralized SGD under Heavy-Tailed Noise: Optimal Convergence Rates and Gradient Clipping

This paper analyzes the impact of gradient clipping and normalization on decentralized non-convex optimization under heavy-tailed noise.

Open in the desk

Coverage

• Decentralized SGD under Heavy-Tailed Noise: Optimal Convergence Rates and Gradient Clipping arXiv: new preprints (security, AI, physics) · tier C · clear net · 2026-10-07 17:57 UTC

What this site indexes