Combining genetic algorithms with squeaky-wheel optimization
Justin Terada, Hoa Vo, David Joslin · 2006
The AI optimization algorithm called "Squeaky-Wheel Optimization" (SWO) has proven very effective in a variety of real-world applications. Although the ideas behind SWO are more closely tied to those of local search such as hill-climbing, in some ways SWO can be thought of as an evolutionary algorithm. From that point of view SWO makes a number of design decisions that are at odds with the conventional wisdom of evolutionary algorithms, but not for any clear reasons. This suggests the possibility of improving on SWO by incorporating aspects of Genetic Algorithms that are known to be effective. We compare several algorithm variants on a set of constrained optimization benchmarks, and present some preliminary results suggesting that combining ideas from SWO with a more standard GA approach yields some significant improvements over both.