ASQ uses an input-adaptive step size for activation quantization and POST uses square-root-of-two levels for weights, reporting 4-bit accuracy near or above full precision on ImageNet.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Precision Neural Network Quantization via Learnable Adaptive Modules
ASQ uses an input-adaptive step size for activation quantization and POST uses square-root-of-two levels for weights, reporting 4-bit accuracy near or above full precision on ImageNet.