A Study on Neural Networks Using Taylor Series Expansion of Sigmoid Activation Function

摘要

The use of microcontroller in neural network realizations is cheaper than those specific neural chips. However, realization of complicated mathematical operations such as sigmoid activation function is difficult via general microcontrollers. On the other hand, it is possible to make approximation to the sigmoid activation function. In this study, Taylor series expansions up to nine terms are used to realize sigmoid activation function. The neural network (NN) structures with Taylor series expansions of sigmoid activation function are used for the concentration estimation of Toluene gas from the trend of the transient sensor responses. The Quartz Crystal Microbalance (QCM) type sensors were used as gas sensors. The appropriateness of the NNs for the gas concentration determination inside the sensor response time is observed with five different terms of Taylor series expansion.