What is the criteria for selecting the range in normalization? Generally the range is 0 to 1. But I have read papers where different ranges are used Say 1 to 7.
I want to know the logic behind this.
 A: If by "normalization" you mean "scaling" than the idea behind this procedure is to ensure that all your variables lie in the same range and have the same importance for the model. 
For example, if you measure your distance in feet and the velocity in m/h - the influence that distance will have would be roughly 0.3 of the influence that the velocity will have.
The common practice for scaling is to make all the features lie in range of [0, 1], but it doesn't really matter in terms of performance, (i.e. you could make all your variables range from 100 to 200 and the performance would be the same), it's only important to scale all of your variables (by all I mean numeric variables off course).
Hope that helps.
Cheers.
A: If your output activation function has a range of [0,1], then obviously you must ensure that the target values lie within that range. But it is generally better to choose an output activation function suited to the distribution of the targets than to force your data to conform to the output activation function.
