Wednesday, May 22
10:00 AM-12:00 PM
It is standard practice in nonlinear programming to exploit firstorder and secondorder information about the objective function and the constraints. This information is typically used to construct the Taylor series approximations that are required by globalized quasiNewton methods. When it works, this strategy is extremely effective, as evidenced by the popularity of quasiNewton methods and the elegant convergence theory that supports them.
Beyond Taylor Series Approxi-mations: The Use of Alternative Models in Nonlinear Programming
But what can be done when, as often occurs in practice, the standard techniques do not apply? What if the objective function is not differentiable, or reliable derivatives are not available? The speakers will examine several promising ideas for nonstandard modeling strategies that address these issues.
Organizer: Virginia Torczon,
College of William and Mary
- Global Modeling for Optimization
- Paul Frank, Boeing Information and Support Services
- An Approach to Derivative-Free Optimization
- Andrew R. Conn, IBM T.J. Watson Research Center
- Managing Approximation Models in Optimization
- J.E. Dennis, Jr., Rice University
- Local Quadratic Models in Stochastic Optimization
- Michael W. Trosset, University of Arizona