Next: The local optimizer
Up: From Genetic Algorithms To
Previous: How does nature
This chapter will explain the application of the second key idea as the
backbone of the local optimization algorithm:
- Adjust the size of probing steps to suit the local nature of the
terrain, shrinking when probes do poorly and growing when probes do well.
In the previous chapter we concluded by introducing the idea of moving
from a coarse-grained search to a fine-grained search. In this
chapter I will demonstrate how this can be achieved by using a
schedule of step sizes that start with large moves, then gradually
shrink down to smaller ones. Instead of using a pre-determined
schedule, the algorithm will adapt its own step size to the landscape
according to the following principle: Expand when making
progress, shrink when stuck. We will see how this contributes to
both the efficiency and the robustness of the optimization. The next
chapter will illustrate how we can further improve the efficiency by
adding a small memory of recent moves.
Deniz Yuret
Tue Apr 1 21:38:29 EST 1997