REXTRA claims O(1/k) convergence for decentralized manifold optimization with single-round iterate communication and constant step size, but the key contraction lemma is false as stated.
Orthogonal weight normalization: Solution to optimization over multiple dependent Stiefel manifolds in deep neural networks
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
-
Riemannian EXTRA: Communication-efficient decentralized optimization over compact submanifolds with data heterogeneity
REXTRA claims O(1/k) convergence for decentralized manifold optimization with single-round iterate communication and constant step size, but the key contraction lemma is false as stated.