Normalization of data in a neural network

1

I'm studying the use of neural networks and learned that input data should be normalized so that the activation functions do not operate in saturation intervals, it is common for the normalization interval to be [-1, + 1]. However, to normalize the input data I must know the range of the same. My question is: If I do not hold the range of these data, how can I normalize them?

    
asked by anonymous 28.11.2017 / 13:30

1 answer

1

If you have the data use the largest and smallest as range, or check the largest and smallest and from them stipulate a range that you consider acceptable.

From the moment you have the range you can normalize with the formula:

    
17.01.2018 / 19:54