A clipped-consensus distributed gradient method provably converges in mean square to a tunable O(a^2) neighborhood of the optimum under infinite-variance communication noise, for heterogeneous strongly convex costs.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
math.OC 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Distributed gradient methods under heavy-tailed communication noise
A clipped-consensus distributed gradient method provably converges in mean square to a tunable O(a^2) neighborhood of the optimum under infinite-variance communication noise, for heterogeneous strongly convex costs.