# Notes theory ML LDA is used to carve up multidimensional space. PCA is used to collapse multidimensional space. If you have few data, this is unstable, and you get overfitted projections towards your data points, i.e. a singular within-class covariance matrix.