why do most algorithm only take number between -1 and 1 as input or output?

[hsunteik]

Why do algorithm like Perlin noise and neural network and many more algorithm must have the input and output to be between -1 and 1,is there a reason for that?
In other words,why do value have to be normalized in algorithm like neural network?
Can I use actual number ,instead of number between -1 and 1?

Second question:
Is gradient interpolation/lerp a must in Perlin noise and simplex noise?
What does it mean?

Third question(not very important):
How do I do infinite terrain using chunk?