... Karhunen¹

This research has been funded by the European Commission project BLISS, and the Finnish Center of Excellence Programme (2000-2005) under the project New Information Processing Principles.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... structure²

By structure, we mean the sizes of the layers and the connections between the nodes. In principle, we could allow any directed acyclic graph connecting the latent and observed variables.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... dynamic³

In [9], NFA was extended to include a model for the dynamics of the sources. A similar extension for HNFA would lead to hierarchical nonlinear dynamical factor analysis.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... dimensions⁴

Higher dimensional SOMs become quickly intractable due to exponential number of parameters.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.