Finding Patterns that Correspond to Episodes TITLE2
Paul R. Cohen, Norman H. Adams, David B. Hand · 2001
We present two algorithms for elucidating structures in time series. These are unsupervised algorithms; they discover patterns without any knowedge about the episodic structures in the time series data. Yet, these patterns correspond with episodes, at least in an experiment with data from robot episodes. We offer a preliminary explanation for this result based on the idea that episodes persist. If this explanation is correct, then the algorithms are apt to be more generally applicable.