Title:
On the Generative Nature of Prediction
Author(s):
Wolfgang Löhr, Nihat Ay
Paper #:
08-02-004
Date:
Feb. 1, 2008
Abstract:
We propose a notion of predictive models that extends the corresponding notion generally used in computational mechanics. We show that this extension allows for hidden Markov models that are more concise than the corresponding epsilon-machines. Thereby it provides more consistency with related theories such as Jaeger's theory of observable operator models. Furthermore, epsilon-machines turn out to be minimal within the usual context of predictive models.
Facebook
Twitter
Email Newsletter
SFI News Feed
Google+
YouTube
iTunes U