Lab Home | Phone | Search
Center for Nonlinear Studies  Center for Nonlinear Studies
 Home 
 People 
 Current 
 Affiliates 
 Visitors 
 Students 
 Research 
 ICAM-LANL 
 Publications 
 Conferences 
 Workshops 
 Sponsorship 
 Talks 
 Colloquia 
 Colloquia Archive 
 Seminars 
 Postdoc Seminars Archive 
 Quantum Lunch 
 Quantum Lunch Archive 
 CMS Colloquia 
 Q-Mat Seminars 
 Q-Mat Seminars Archive 
 P/T Colloquia 
 Archive 
 Kac Lectures 
 Kac Fellows 
 Dist. Quant. Lecture 
 Ulam Scholar 
 Colloquia 
 
 Jobs 
 Postdocs 
 CNLS Fellowship Application 
 Students 
 Student Program 
 Visitors 
 Description 
 Past Visitors 
 Services 
 General 
 
 History of CNLS 
 
 Maps, Directions 
 CNLS Office 
 T-Division 
 LANL 
 
Wednesday, January 09, 2013
3:00 PM - 4:00 PM
CNLS Conference Room (TA-3, Bldg 1690)

Seminar

The Knowledge Gradient Policy for Optimal Learning

Warren B. Powell
Department of Operations Research and Financial Engineering, Princeton University

There are many applications which require collecting information, where the time or cost required to make a measurement may be high. A measurement may require running an expensive simulation, testing a molecular compound in a lab, estimating the presence of a disease in the population, or field testing a market price or business policy in the market place. There are elegant, optimal techniques for very specialized problems such as multi-armed bandit problems, and a host of heuristics and techniques developed for specialized problem classes. The knowledge gradient is a technique that guides measurement decisions using what might be described as classical steepest ascent which requires finding the expected value of a single measurement. This technique is myopically optimal, and is also asymptotically optimal, with strong supporting evidence for problems with finite budgets. The appeal of the method is its generality, allowing it to address problems that have been previously viewed as belonging to completely separate communities. However, it introduces a specific computational challenge which has to be overcome before it can be used for a particular application. The idea will be illustrated on discrete choice problems with correlated beliefs, scalar problems (e.g. optimizing prices), continuous multidimensional problems (e.g. finding the best set of parameters to optimize a simulation), and drug discovery. This work is joint with Peter Frazier and Ilya Ryzhov.

Host: Frank Alexander, fja@lanl.gov, 665-4518