next up previous contents
Next: Hierarchical Nonlinear Factor Analysis Up: Nonlinear Models Previous: Nonlinear Factor Analysis

Nonlinear Dynamical Factor Analysis

The NFA model can be extended to model the nonlinear dynamics of the factors [20,65,63,64]

 
x(t) = f(s(t)) + n(t) (2.15)
s(t) = g(s(t-1)) + m(t) , (2.16)

where the m(t) is called process noise or innovation process. The mapping g modelling the dynamics is handled somewhat similarly as the mapping f. Experiments [65,63] have given good results in prediction of artificial time series compared to traditional time series methods.



Tapani Raiko
2001-12-10