Real data
   |
   v
Learn its probabilistic structure
   |
   v
Model Distribution
   |
   +--> evaluate data probability
   |
   +--> generate new samples