Sample set ELM GA-SVM SDAE-SVM 300 0.08 17.32 0.18 1500 0.12 45.13 0.20 3000 0.21 215.04 0.41 TABLE 6: Classification accuracy using SDAE-SVM with PReLU and
sigmoid functions (OA%).
The relationship between HR and workload was fitted in linear, quadratic, exponential, and
sigmoid functions. The best mathematical fitting was considered to be the one with the lowest residual mean square.
The
Sigmoid function has very important properties, including the following (see [18]):
The estimated parameters from the
sigmoid function fitted to the TEP recruitment input-output curves are shown in Tables 1 and 2.
The input-output nonlinear
sigmoid functions and wavelet networks, saturation, one-dimension polynomial, and piecewise functions are applied with 50 iterations.
We are adopted by the neural network with a hidden layer of three layer forward network, hidden layer and output layer nodes with a standard of
Sigmoid function. General approximation theorem shows that the feedforward neural network has a hidden layer to achieve any approximation.
The controlled accuracy second order approximation of
sigmoid function was implemented on single DSP slice with maximum allowable 1 % error [9].