High precision expansion of multi-layer neural networks
Qianyi Wang, T. Aoyama, Umpei Nagashima · 2003
We researched high precision multi-layer neural networks for function fittings. We proposed an expansion of neural networks, where the neuron functions (sigmoid-function, sine function series, and variable dumping factors) are used. The expansion improves the fitting precision of the original neural networks in 1000 times.