ruth@utstat.uucp (Ruth Croxford) (09/27/88)
Topic: Principal Curves Speaker: Trevor Hastie Statistics and Data Analysis Group AT&T Bell Laboratories Murray Hill, New Jersey Date: Thursday, September 29, 1988, 4:00 p.m. Place: Sidney Smith Hall, room 2110, 100 St George St, University of Toronto Abstract: The first linear principal component of a p-dimensional point cloud provides the line that minimizes the distance from the points to itself. It is a one dimensional summary of the data, but often a linear summary is inadequate. Principal curves are smooth one dimensional curves that pass through the middle of the data. They minimize the distance from the points, a non-linear summary of the data. In practice the curves are estimated using smoothers and so their shape is suggested by the data. After a brief introduction, I will show a short movie that illustrates the curves and demonstrates an algorithm for fitting them. This will be followed by some technical definitions and results. Finally, I will briefly describe several areas of practical importance where the techniques have been successfully applied. ---------------------- Coffee and tea will be served in the Delury Lounge (SS6006) at 3:30 p.m.