Incremental Gradient Descent has worst-case convergence gaps in the small epoch regime that become exponentially bad with nonconvex components, though a carefully chosen fixed permutation can still outperform with-replacement SGD.
Tighter lower bounds for shuffling SGD : Random permutations and beyond
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Incremental Gradient Descent with Small Epoch Counts is Surprisingly Slow on Ill-Conditioned Problems
Incremental Gradient Descent has worst-case convergence gaps in the small epoch regime that become exponentially bad with nonconvex components, though a carefully chosen fixed permutation can still outperform with-replacement SGD.