Abstract Details
Activity Number:
|
69
|
Type:
|
Topic Contributed
|
Date/Time:
|
Sunday, August 9, 2015 : 4:00 PM to 5:50 PM
|
Sponsor:
|
Section on Bayesian Statistical Science
|
Abstract #315692
|
View Presentation
|
Title:
|
Speeding Up Neighborhood Searches in Local Gaussian Process Fitting of Large-Scale Computer Experiments
|
Author(s):
|
Ben Haaland* and Chih-Li Sung
|
Companies:
|
Georgia Tech, ISyE and Georgia Tech, ISyE
|
Keywords:
|
computer experiment ;
local Gaussian process ;
subsampling ;
big data
|
Abstract:
|
Local Gaussian process fitting is a promising tool for predicting the output of very large-scale computer experiments. Predictions at distinct locations are based on distinct sub-samples of the full dataset, making the predictive task fully parallelizable. Efficient and scalable selection of the sub-sample to be used for prediction at a particular location remains a potential bottleneck. While it can be shown that a K-nearest-neighbor approach is not optimal, here we show that data locations which are sufficiently distant from the location of interest and the current sub-sample have arbitrarily small potential to reduce the predictive variance. Further refinements of the set of candidate data locations are possible via a projection into (residual) feature space. This feature space reveals a balance between nearby data locations and data locations along one-dimensional slow manifolds through the location of interest. An efficient and scalable sub-sample selection algorithm is presented.
|
Authors who are presenting talks have a * after their name.
Back to the full JSM 2015 program
|
For program information, contact the JSM Registration Department or phone (888) 231-3473.
For Professional Development information, contact the Education Department.
The views expressed here are those of the individual authors and not necessarily those of the JSM sponsors, their officers, or their staff.
2015 JSM Online Program Home
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.