Back To Schedule
Thursday, August 22 • 1:00pm - 1:45pm
Shapelets via Procrustes Tangent Distance matching

Log in to save this to your schedule, view media, leave feedback and see who's attending!

Feedback form is now closed.
Shapelets are compact abstracted feature descriptors mined from time series. They have been used for qualitative and high school cross-correlation of series, and for classifying phenomena and behaviors in evidence within series, particularly if the record is of human activity. Traditionally these are devised by hand, with domain knowledge, and then procedures like variable selection used to winnow and tune them. Automatic discovery has been explored, using techniques like Dynamic Time Warping, but these can be expensive. Matching using Procrustes Tangent Distance (Dryden, Mardia, Kent, Rohlf, 1993, 1994, 1999) is introduced here, and libraries of features are built by auto-correlating series using this operator. Libraries from different series can then be compared to discern similarities and generalize. Applications to electricity consumption series and to hydrological stream flow are used to illustrate.

avatar for Jan Galkowski

Jan Galkowski

Statistician, Quantitative Engineer, Westwood Statistical Studios
Professionally, I do Internet sociology at a major Internet company in the Kendall Square area. I also volunteer statistical skills to local governments and non-profits, helping to promote sustainability and assess spatiotemporal risks. My technical interests are computational statistical... Read More →

Thursday August 22, 2019 1:00pm - 1:45pm
Room 408 595 Commonwealth Avenue, Boston, MA 02215, USA

Attendees (15)