Tuesday, May 21
2:45-4:25 PM
Esquinalt
MS24
Recent Developments in Non-Derivative Optimization Methods
During the past five years, there has been renewed interest in non derivative methods, particularly for problems in which function evaluations are so expensive or complicated that exact or approximate derivatives cannot be calculated at reasonable cost. The need for research is heightened because direct search methods, notably the 1965 Nelder Mead method and variants, are widely used in practice despite serious deficiencies in theory and performance. The three speakers will cover a range of very recent theoretical and numerical research in non derivative optimization: trust region methods using quadratic interpolation; a new class of direct search methods based on `fortified descent'; and convergence properties of the original Nelder Mead method.
Organizer: Margaret H. Wright,
Bell Laboratories
- Trust Region Methods that Employ Quadratic Interpolation to the Objective Function
- Michael J.D. Powell, University of Cambridge, United Kingdom
- Direct Search Methods with Fortified Descent: An Update
- Paul Y. Tseng, University of Washington
- Theoretical and Numerical Properties of the Nelder-Mead Method
- Margaret H. Wright, Organizer
MEM, 3/18/96