This is a subject I struggled with the first time I took it. Ironically, this was the engineering version of it. It wasn't until I took the rigorous, axiomatic version that everything clicked.

Suitable linear transformations for preliminary data analysis are available when data consist solely of continuous variables or solely of binary variables. When mixtures of variables are observed, ...

Journal of the Royal Statistical Society. Series B (Statistical Methodology), Vol. 76, No. 2 (MARCH 2014), pp. 349-372 (24 pages) The paper considers in the high dimensional setting a canonical ...