Theoretical Investigation of TD- and PCA-Based Unsupervised FE

In this chapter, we discuss TD- and PCA-based unsupervised FE from the theoretical point of views. At first, we show that those can be regarded to be projection pursuit, an old fashioned machine learning method. Next, we figure out why the histogram of P values cannot coincide with the null hypothesis and show that the optimization of standard deviation can resolve this problem.