 |
|
|
|
2004 MCS Divisional Seminars & Colloquia |
|
An Efficient Sampling Approach to Multi-objective
|
|
| Abstract | The technique of Automatic differentiation allows the computation of
exact Jacobian-vector, vector-Jacobian and Hessian-vector products. This information allows the implementation of a trust region algorithm
that is completely matrix free; that is, neither the exact Hessian and Jacobian nor approximation thereof have to be stored. Therefore, the
presented algorithm applies iterative solvers to compute the normal and the tangential steps instead of forming and factoring the derivative
matrices directly. First numerical results for a part of the CUTE test sets are shown. Furthermore, possible extentions to optimization
problems involving inequality constraints are discussed.
|
| [MCS | Research | Resources | People | Collaboration | Software | Publications | Information] | |||
|
