Given a modest number of evaluations of the objective (a ``budget'')
produce
- 1.
- a better ``picture'' of the behavior of the objective in the
region(s) of interest and
- 2.
- do so in a way that is guaranteed--in theory--to converge to
at least a local stationary point of the function.
With a limited number of objective evaluations, it may not be possible
to ensure convergence to a stationary point, but satisfying the
conditions required by the theory provides some assurance that this is
a sensible approach to the design optimization problem.
Next: Fundamental Idea:
Previous: Consequences:
Virginia Torczon
6/24/1998