Binary normalized layers with all parameters in {0,1} reach near-32-bit accuracy on Food-101 and WikiText-103, with some caveats.
Quantized neural networks: Training neural networks with low precision weights and activations
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
1 bit is all we need: binary normalized neural networks
Binary normalized layers with all parameters in {0,1} reach near-32-bit accuracy on Food-101 and WikiText-103, with some caveats.