Principal Component Analysis
Sometimes a few dimensions explain most of the variability
Here 96 dimensional feature vectors are generated by Gaussian distributions
Then they are reduced to 3 dimensions using Principal Componenet Analysis
Then they are visualized and classified