In this video on machine learning by Andrew Ng, called "Features and Polynomial Regression", at time 4:34, he mentions the possibility of feature normalization in polynomial regression. By which he means, I believe, applying an affine transformation to the feature vector $(1, x, x^2, \dotsc, x^{n-1})$ so that the variances of the $x^i$ are closer to one another.
Does anyone know of a reference on this practice? I'd like to investigate it and its effects. (That is, the specific application of feature normalization to polynomial regression. Not just feature normalization in general.)