"Bias" parameter in Neural Network

What is your definition of "bias" in this case?  For the NN operator, the "bias" returned for each node is effectively the intercept term applied to the equation form of the fit curve at each perceptron.  There is no corresponding "bias" at the overall model level, just as there is no set of single coefficients for the input variables (each node has its own set of coefficients).  So what exactly are you trying to measure at the level of the overall model?

Hi 
I know what you mean. But in some published data using other software (Matlab etc.), it has been reported just one Bias for training and/or testing process. So, it does confusing me! Please see the attached file.

Best

Hello @mb223223

I got your point. As @Telcontar120 said the bias is for each node in an NN. What MATLAB does is it will give bias of each node in the form of a vector-based on a command getwb. I am not sure, but a guess as I didn't read the paper you mentioned. I think they might have done some average or other operation to get the training bias and testing bias.

You can also get the bias of each node in Rapidminer. Once the mode is run, it will be in the description of the model.