[PDF][PDF] Learning the best combination of solvers in a distributed global optimization environment
Proceedings of Advances in Global Optimization: Methods and Applications (AGO�…, 2007•esa.int
… Our aim is not to find the best solver or to find better solution to the problems taken into
account, but to use a fixed setup of the solvers, to allow a fixed number of objective function
evaluation and to see if the proposed cooperative strategies are able to improve the
performances. … ▶ We find that this cooperative solver outperforms the stand alone ones
already if a random probability is given to assign to a client a particular solver. ▶ We also
find that a further improvement is possible by letting the server learn which one is the most�…
account, but to use a fixed setup of the solvers, to allow a fixed number of objective function
evaluation and to see if the proposed cooperative strategies are able to improve the
performances. … ▶ We find that this cooperative solver outperforms the stand alone ones
already if a random probability is given to assign to a client a particular solver. ▶ We also
find that a further improvement is possible by letting the server learn which one is the most�…
Rule C. Let μI the mean fitness value of the selected subpopulation before the run and μO after the run of the selected solver. Put the difference μO− μI to the stack. We refer to this rule as CO-μ-valueofa (eg CO-μ-0.1).
esa.int
Showing the best result for this search. See all results