next up previous
Next: Introduction Up: Workpage of Harri Valpola

Unsupervised Learning of Nonlinear Dynamic State-Space Models

Harri Valpola
Neural Networks Research Centre
Helsinki University of Technology
P.O.Box 5400, 02015 HUT, Finland


This technical report describes how the nonlinear factor analysis algorithm introduced in [4] can be extended by taking into account the dynamics of the factors. The resulting algorithm represents the observations as having been generated by a nonlinear dynamical system. The internal states and the nonlinear mappings describing the nonlinear system are assumed to be unknown and they are estimated from the observations. Experiments with artificial data verify that the algorithm is able to reconstruct the dynamics of the underlying process which has generated the observations.


Harri Valpola