16-bit log-based training achieves classification accuracy within approximately 1% of floating-point baselines on common datasets by replacing multiplications with approximate log-domain additions.
Thus, v ← →V = (V,sv) (1a) V = log2 (|v|) (1b) sv = sign(v) (1c) where sign(v) = 1 ifv > 0 and 0 otherwise
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2019 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Neural Network Training with Approximate Logarithmic Computations
16-bit log-based training achieves classification accuracy within approximately 1% of floating-point baselines on common datasets by replacing multiplications with approximate log-domain additions.