Feature Space
Is a coordinate system (often mapped to from sample space) in which correlation is reduced (variation is increased) as much as possible.
The goal of feature-space is essentially to find a set of axes in which correlation is as small as possible. Note that correlation is bad. Information is repeated across different feature/dimensions and when training a system it needs to detect and disregard this overlap lest dealing with unintended biases. We want our dimensions to be generally independent of each other.