ElasticZO uses zeroth-order gradients for most layers and backpropagation only for the last few, improving accuracy over pure ZO training with tiny memory overhead, and its INT8 variant runs integer-only.
Hero III, and Pramod K
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
-
ElasticZO: A Memory-Efficient On-Device Learning with Combined Zeroth- and First-Order Optimization
ElasticZO uses zeroth-order gradients for most layers and backpropagation only for the last few, improving accuracy over pure ZO training with tiny memory overhead, and its INT8 variant runs integer-only.