"Measurement Selection and Receiver Operating Characteristics"

Information Science and Technology Center Seminar Series

Hosted by the Information Science and Technology Center (ISTC)

Presented by:  Andy Fraser, Los Alamos National Laboratory

Abstract:  I will summarize what I've recently learned about evaluating evidence or measurement options. In each of the two examples that I describe, one must make a binary decision. Before making that decision one can choose between two evidence options.

I've found two ways to quantitatively compare evidence options. One is Bayesian and requires priors and the costs of classification errors as inputs. The other is asymptotic and is only valid in the limit of a large number of measurements that are iid given the class.  Both approaches are classical. In learning this material, I've used both \emph {Elementary Decision Theory} by Chernoff and Moses and \emph {Elements of Information Theory} by Cover and Thomas.

Biography:  Andy Fraser is a Scientist in the Space and Remote Sensing Sciences division of Los Alamos National Laboratory, where he uses stochastic models in his work on signal analysis. Before coming to the Lab, he was on the faculty of Portland State University in both the Systems Science Program and the Electrical and Computer Engineering department.  He earned a Ph.D. in Physics from the University of Texas at Austin for work on the application of ideas from Information Theory to measurements of chaotic dynamics. A new flyer listing SIAM's best selling titles includes Andy's recent book: "Hidden Markov Models and Dynamical Systems."

Contact the technical host Garrett Kenyon, gkenyon@lanl.gov, 667-1900, for further information.