Neural network input normalization by mean / standart deviation : how to do it? -

i consufed on how normalize inputs / outputs regression neural network using (gaussian normalization ? ) mean & standart deviation normalization technique :

most importantly, normalize data ?

let me explain :

let's have these training data on 2 input neurons, 2 hidden neurons , 1 output neuron:

[input1 : 10][input2: 5] [input1:  30][input2: 255] 

do normalize column(neuron), or inputs data ? mean input neuron 1 =

(10+30)/2  

or

(10+30+5+255)/4 ?  

try both weird result using typical xor example (only 1s , 0s in traning data), loosing great accuracy when normalizing.

normalization keep each dimension of input data in range should done in column. there several ways normalization. example, linear normalization: it's common easiest method , used when data centered. it's counted (v-vmin)/(vmax-v). , gaussian normalization counted (v-vavg)/std.