Research Interests: Predictive models of time series
Assume that you are given a time series and your task is to build
a model from the observations of the past that you have available. This
problem occurs not only in time series analysis, but also for any
autonomous, adaptive agent. Animals and humans are presumably
quite good at this task, which they need to solve for survival.
Intuitively, a good model is one that predicts the data one hasn't
encountered yet, i.e. the future. Furthermore, if two models acheive
the same in terms of prediction, but one of the models is simpler than
the other one, we prefer the simpler model.
This intuition can be made precise, using information theory. One can
derive and study the class of optimally predictive models, and a
process's causal compressibility.