[Ml-stat-talks] Talk on surrogate global optimization
powell at princeton.edu
Mon May 12 12:28:02 EDT 2014
To the ML community - surrogate global optimization is a nice mixture of
machine learning and optimization (the problem setting is the same that
motivates bandit problems).
Christine is giving the Andlinger highlight talk this afternoon, but the
talk tomorrow is an algorithms talk in ORFE.
In Room 101, Sherrerd Hall at 11am on Tuesday, May 13:
Surrogate Global Optimization for Computationally Expensive Objective
Functions with Integer and Continuous Variables with Applications
*Christine A. Shoemaker*
Joseph Ripley Professor, Cornell University
School of Civil and Environmental Engineering
School of Operations Research and Information Engineering
This talk presents a framework for parallel surrogate-based constrained
nonlinear optimization algorithms for computationally expensive objective
functions with continuous and/or integer variables. The algorithms converge
almost surely. The methods are efficient when the objective function or
the constraints are computationally expensive. “Efficient” indicates an
algorithm gets solutions with relatively few evaluations of the objective
function, which is often a computationally intensive simulation model (e.g.
minutes or hours per simulation). The framework is designed for problems
with a possibly multi-modal objective function that is black-box so that
derivatives and the number of local minima are not known. The algorithm
results are numerically compared to other pure and mixed integer and
continuous variable algorithms on test problems. Applications to a
continuous variable groundwater transport problem and to an integer
optimization problem arising in watershed phosphorous pollution management
will be presented.
Shoemaker’s algorithms address local and global continuous and integer
optimization, stochastic optimal control, and uncertainty quantification
problems. In her recent research algorithm efficiency is improved with the
use of surrogate response surfaces iteratively built during the research
process and with intelligent algorithms that effectively utilize parallel
and distributed computing. Her applications areas include physical and
biological groundwater remediation, carbon sequestration, pesticide
management, ecology, and calibration of climate and watershed models.
Algorithms that are efficient because they require relatively few
simulations are essential for doing calibration and uncertainty analysis on
computationally expensive engineering simulation models. Professor Shoemaker is
a member of the National Academy of Engineering and is a Fellow in AGU,
SIAM, INFORMS and Distinguished Member of ASCE.
*------------------------------Warren B. Powell*
*Professor, Department of Operations Research and Financial Engineering*
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Ml-stat-talks