Simulated data is sometimes used in machine learning when there is either insuffcient real data or it is hard the gather for some reason. An example of the latter is trainng AI for an autonomous car; it is useful to have examples of people stepping off the curb in front of the car, but not a good idea to gather the information for real, so that VR and games engines are used to create realistic scenarios with lots of variations. This example is were the simulated data is created from a model. In other cases real data is used to prime the process, for example, generating lots of small perturbations from each real data item.
Used in Chap. 20: page 318