Instructor: James Gentle
Many methods of statistical analysis can be formulated as optimization problems. These include maximization of a likelihood function for estimation, minimization of residuals for fitting models, optimization of experimental designs, and minimization of risk or expected loss in general problems of making decisions under uncertainity.
This course covers various methods of optimization, including the traditional descent methods under assumptions of smoothness of functions and methods for combinatorial optimization. In addition to the usual methods of numerical analysis, Monte Carlo methods and heuristic methods will be discussed.
The numerical methods will be motivated by problems in statistical data analysis.
Some knowledge of statistical theory and methods (roughly equivalent to STAT 554 and STAT 652) and an introduction to computers are prerequisites.
Performance in the class will be evaluated based on
Each student will prepare a Web page for presentation of the project and for some of the smaller assignments.
There is no text for the course. Notes developed by the instructor will be given out from time to time, and some will be put on the net. Notes
Links to some useful Web sites will also be provided.
For the project, it will be necessary to have access to statistical research journals. The Journal of the American Statistical Association, available in Fenwick Library and other places, is sufficient for this purpose.
The most important WWW repository of statistical stuff (datasets, programs, general information, connection to other sites, etc.) is StatLib Index at Carnegie Mellon.
James Gentle, jgentle@gmu.edu