No-replacement sampling yields a bounded, smoother batch-size update formula for variance-controlled SGD, at the cost of no new convergence proof and no experiments.
, year 2016
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
math.OC 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
On the Batch Size Selection in Stochastic Gradient Methods Using No-Replacement Sampling
No-replacement sampling yields a bounded, smoother batch-size update formula for variance-controlled SGD, at the cost of no new convergence proof and no experiments.