A mixed hardware-software approach to flexible artificial neural network training on FPGA
SAMOS'09 Proceedings of the 9th international conference on Systems, architectures, modeling and simulation
| Hi-index | 0.00 |
The paper deals with implementing nonlinear sigmoid-like squashing functions in FPGA based neural networks. Five piecewise linear (PWL) approximation techniques and a look-up table are compared. It is examined their performance in terms of recall delay, area requirements, and accuracy in two FPGA neural networks.