Posit multiplier template in FloPoCo enables neural network training and inference with reduced-precision posits, matching FP accuracy on MNIST but not CIFAR-10.
Design of power-efficient fpga convolutional cores with approximate log multiplier
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2019 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Template-Based Posit Multiplication for Training and Inferring in Neural Networks
Posit multiplier template in FloPoCo enables neural network training and inference with reduced-precision posits, matching FP accuracy on MNIST but not CIFAR-10.