Incrementally binarizing a Gated Recurrent Unit network preserves enough source-separation quality that a fully binary single-layer model outperforms larger full-precision fully connected networks.
The training is done in two roun ds, first in a weight compressed network and then in an incrementa lly bitwise version with the same topology
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
eess.AS 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Incremental Binarization On Recurrent Neural Networks For Single-Channel Source Separation
Incrementally binarizing a Gated Recurrent Unit network preserves enough source-separation quality that a fully binary single-layer model outperforms larger full-precision fully connected networks.