Seminar on Computational Learning and Adaptation




Innovations in Local Modeling for Time Series Prediction


James McNames
Information Systems Laboratory
Stanford University
mcnames@teleport.com



Unlike most time series studied in statistics, chaotic time series generally contain very little noise, are highly nonlinear, and are very long (1,000-100,000 points). For prediction, researchers have often used nonlinear regression algorithms such as radial basis functions, neural networks, MARS, and k-nearest neighbors regression (loess). However, there are several important differences between the regression problem and the time series prediction problem that have been mostly overlooked. A new local modeling algorithm will be described that incorporates these differences to increase accuracy and reduce computation.


Date: Wed., May 27; Time: 4:15-5:30PM; Place: Cordura 100


The goal of this seminar is to increase communication among local researchers with interests in computational approaches to learning and adaptation. If you would like to be added to (or removed from) the mailing list, or if you are interested in giving a talk in the seminar, please send email to iba@isle.org.


Return to seminar schedule.