SAMCO: Surrogate-Assisted Multi-Criteria Optimization

In many optimization tasks, different conflicting aims need to be considered. Due to changing customer groups and needs, in most cases no a priori preference information is available. Hence, we are interested in finding a set of solutions that reflects the optimal trade-offs among the conflicting optimization criteria. This kind of problem, we call a (set-based) multi-criteria optimization (MCO) problem.

In real-world applications, a lot MCO problems do arise. Those problems go beyond the consideration of multiple independent objectives, such as quality and cost.  For instance, multidisciplinary and robust design decisions face multiple views on a single objective function.

Beginning with the seminal Efficient Global Optimization paper of Jones et al. in 1998, sequential surrogate-assisted optimizers became accepted for solving problems with expensive black box objective functions. In those sequential approaches, the surrogate model estimates the response surface of the problem by means of an initial experimental design. This surrogate model allows the next solution to be selected for evaluation based on a suitable figure of merit.

With our SAMCO workshop, we aim at bridging the research in both topics, set-based multi-criteria optimization and surrogate-assisted optimization. The intersection of these topics is an emerging research area, but the research in the field is still very heterogeneous and distributed over different countries and faculties. Our workshop thus brings together researchers from all backgrounds and allows new concepts connected to SAMCO to be discussed under a holistic perspective. In particular, we plan five key aspects of SAMCO to be addressed:

1) Overview of existing libraries and approaches

2) Beyond one model per objective function: higher-level surrogate approaches

3) Ensembles of surrogate models

4) Benchmarking

5) Theoretical aspects

More details and the results of the workshop can be found on our website: http://samco.gforge.inria.fr