Approximation of the Pareto Frontier

Tomi Haanpää

Most of the current Pareto frontier approximation models 
are for convex multiobjective optimization problems. 
In the convex case, one can construct very accurate
approximation of the frontier, but the situation is
totally different, if the problem is nonconvex.
In a nonconvex case, one can have extremely badly behaving 
Pareto frontiers. Therefore the approximation model
fitting process is not an easy task to do. 

One solution for the nonconvex case is an approximation of 
the image of the objective function and a classifier based
approximation of the frontier by using the approximation
of the image is illustrated.