| |
Abstract:
We present a new technique for time series analysis based on
dynamic probabilistic networks. In this approach, the observed data
are modeled in terms of unobserved, mutually independent factors,
as in Independent Factor Analysis (IFA) recently introduced by
(Attias 1999). However, unlike IFA, the factors are not i.i.d.;
each factor has its own temporal statistical characteristics. We
derive a family of EM algorithms that learn the structure of the
underlying factors and their relation to the data, and demonstrate
superior performance compared to IFA. Model selection issues, in
particular inferring the optimal number of factors, are also
addressed.
|
