Does weights and bias received from ANN are also normalized???

Question

Radek 2016-4-20

1
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/280022-does-weights-and-bias-received-from-ann-are-also-normalized

评论： Greg Heath 2017-2-1

I have gained the weights and bias from ANN and inserted them into a simple mathematical formulea function derived from the ANN architecture and does not give good results. So, If the gained weights from the ANN are also automatically normalized? If yes, so how to get the real them value?

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Greg Heath 2016-4-20

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/280022-does-weights-and-bias-received-from-ann-are-also-normalized#answer_218729

Typically, the NN training algorithms use the MAPMINMAX transformation to obtain input and target variables in [ -1 1 ] and to reverse the target transformation to get unscaled outputs.

Therefore, either kill the normalization/unnormalization in the net function and use normalized variables as inputs and targets

OR

include the normalization/unnormalization into your weight formulas.

Do both, compare, and choose the one you prefer.

Hope this helps.

Thank you for formally accepting my answer

Greg

3 个评论
显示 1更早的评论隐藏 1更早的评论

Greg Heath 2017-1-31

Something like

net.processFunctions = ...

Greg Heath 2017-2-1

However, I do not recommend killing normalization. Normalization is used to assure that sigmoidal functions are not saturated.

Since normalization and un-normalization are automatic defaults, you never have to worry about them unless you try to run the programs in a non-MATLAB setting.

请先登录，再进行评论。