# Neural networks - explanation of normalization of data

I'm beginning with neural networks. Currently I'm struggling with fitting to a data-set, which has a large variance both in input variables.

It looks like this:

I1   I2   I3   O1   O2   O3
0.2  0.3  1500 1200 1100 1300
0.22 0.15 1200 1250 1110 1290


I think, that normalization of the data could help me with the numerical problems. But, I'm not sure, if I understand the very basic principle of normalization correctly.

My understanding is, that I'll simply transform all my input values to the range [0,1], which will reduce the data variance. But, should I also normalize my outputs in the same way? Or should I normalize my inputs column-by-column and not altogether?

I've looked at the following questions, but I didn't find an answer to such a basic question: