Biased-DMT achieves linear speedup and improved convergence in nonconvex decentralized stochastic optimization despite biased gradients, eliminating structural heterogeneity error for absolute bias cases.
On the linear speedup analysis of communication efficient momentum SGD for distributed nonconvex optimization
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
math.OC 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Improved Convergence for Decentralized Stochastic Optimization with Biased Gradients
Biased-DMT achieves linear speedup and improved convergence in nonconvex decentralized stochastic optimization despite biased gradients, eliminating structural heterogeneity error for absolute bias cases.