Reflective Random Indexing (RRI) is the name given to the process of training a semantic model in several phases, starting with RandomProjection and then using several TrainingCycles.

See Reflective Random Indexing and indirect inference: a scalable method for discovery of implicit connections. RRI has been used to infer accurate connections between concepts that do not occur together explicitly in text - like LSA, but at a much smaller computational cost.