A time series is a vector of events collected over time. Contrary
to standard random samples, the observations in a time series are
dependent. I develop automated methods for modeling and clustering
time series. This research is supported by NSF.
Clustering: I developed
BCD, a Bayesian model-based
algorithm for clustering categorical time series on the basis of
their dynamics. BCD was extended to clustering continuous time series
and was applied to profiling gene expression data.
Incremental Bayesian segmentation:
The idea is to develop methods for processing a time series on line
and for incrementally recognizing changes in the underlying dynamics.