A large number of objectives leads for most problems to a significant drop in selection pressure.As a consequence,all algorithms,in this setup,tend to approach the behaviour one would expect to see when using a completely random sampling.A main cause of this problem is that any two random solutions become increasingly likely not to be comparable from a Pareto dominance point of view.As a step forward towards addressing this problem,we present a set of coordinates transforms which can be used to control dominance.The same formalism can be used from within an adaptive or dynamic algorithm,with the strength and bias of each transform being varied during the sampling process.
展开▼