Approximation of Markovian models with non-constant parameters

A generalization of the canonical correlation analysis approach has been developed for non-stationary process generated by Markovian models with non-constant parameters. This generalization, is then used to develop two model reduction (approximation) algorithms.