Statistics Seminar - Robert Gramacy

Event Details
  • Date/Time:
    • Thursday November 5, 2015
      10:00 am
  • Location: Advisory Board Room, GC 402
  • Phone:
  • URL:
  • Email:
  • Fee(s):
    N/A
  • Extras:
Contact
No contact information submitted.
Summaries

Summary Sentence: Statistics Seminar - Robert Gramacy

Full Summary: No summary paragraph submitted.

TITLE: Local Gaussian process approximation for large computer experiments

ABSTRACT:

We provide a new approach to approximate emulation of large computer experiments. By focusing expressly on desirable properties of the predictive equations, we derive a family of local sequential design schemes that dynamically define the support of a Gaussian process predictor based on a local subset of the data. We further derive expressions for fast sequential updating of all needed quantities as the local designs are built-up iteratively. Then we show how independent application of our local design strategy across the elements of a vast predictive grid facilitates a trivially parallel implementation. The end result is a global predictor able to take advantage of modern multicore architectures, GPUs, and cluster computing, while at the same time allowing for a non stationary modeling feature as a bonus. We demonstrate our method on examples utilizing designs sized in the tens of thousands to over a million data points.  Comparisons are made to the method of compactly supported covariances, and we present applications to computer model calibration of a radiative shock and the calculation of satellite drag.

Additional Information

In Campus Calendar
No
Groups

H. Milton Stewart School of Industrial and Systems Engineering (ISYE)

Invited Audience
Undergraduate students, Faculty/Staff, Graduate students
Categories
Seminar/Lecture/Colloquium
Keywords
No keywords were submitted.
Status
  • Created By: Anita Race
  • Workflow Status: Published
  • Created On: Oct 30, 2015 - 3:58am
  • Last Updated: Apr 13, 2017 - 5:17pm