Decomposition methods for Pareto frontier approximation in large multi-objective decision problems

Organização: Centro de Estudos Florestais; Centro de Ecologia Aplicada Prof. Baeta Neves
Data: 4 de Fevereiro :: 12h30 às 13h30
Local: Sala PF 1.6

Orador: Alexander V. Lotov, Professor e Investigador no Dorodnicyn Computing Centre of Russian Academy of Sciences, Moscow, Russia.

Resumo da apresentação:
The talk is devoted to new results in the field of Pareto frontier approximation and visualization in multi-objective decision problems. The Interactive Decision Maps (IDM) technique that has successfully been used during several decades is extended to multi-objective problems with a large number of decision variables. The IDM technique applies interactive visualization and animation of the Pareto frontier in the form of decision maps, i.e. collections of the objective tradeoff curves. By studying the decision maps the decision makers learn the feasible combinations of the objective values (the feasible goals) and the objective tradeoffs. This information supports decision maker in identifying the preferred feasible goal, which is specified directly at one of the tradeoff curves. Then, computer constructs the decision associated with the goal. To implement the IDM technique, the Edgeworth-Pareto hull (EPH), that is, the largest set that has the same Pareto frontier as the multi-objective problem under consideration is approximated in advance. In the case of a large number of decision variables, the decomposition methods for EPH approximation are proposed. The general scheme of decomposition methods is based on splitting the system under study into subsystems, approximating the EPH for subsystems and using the approximations for constructing the EPH for the whole system. Several particular methods based on this scheme as well as approximation examples are considered.