A New Challenge: Saving Mark Whitney
In the last blog entry Adding Crossover to CMA-ES we provided further details to the new optimimization algorithm GCMA-ES designed to partly replace search by optimization for space mission design, but also can be applied to hard optimization problems in other areas. As test method we used ESAs GTOP messenger full problem . We observed that ESAs current implementation is flawed, but can easily be fixed by updating the Lambert solver implementation. Even the updated version has a problem: It is quite "easy" to solve for new algorithms. On a powerful single processor machine like the TR2990WX the same optimum is found every few hours. We cannot be sure it is a global optimum, but would prefer a problem where better solutions as the ones currently found are expected, enabling us to measure further improvements of real parameter optimization algorithms. If you know of such a problem, specially if it is not related to space mission planning, please comment. Extending ESAs...