The paper extends implicit graduated optimization, which views SGD noise as smoothing, to momentum-based SGD, gives a convergence analysis, and reports empirical gains on image classification, but the analysis has a load-bearing mismatch with the pseudocode.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Explicit and Implicit Graduated Optimization in Deep Neural Networks
The paper extends implicit graduated optimization, which views SGD noise as smoothing, to momentum-based SGD, gives a convergence analysis, and reports empirical gains on image classification, but the analysis has a load-bearing mismatch with the pseudocode.