The primary concepts behind genetic algorithms are biological evolution and representation:
The genetic algorithm itself consists of multiple stages:
Watch out for a few details in practice, such as premature convergence and slow convergence. These issues can be resolved by tuning the parameters appropriately. Despite not being particularly efficient, genetic algorithms are an extremely robust and flexible optimization strategy. Chapter 34, "Adaptive Defensive Strategies with Genetic Algorithms," applies genetic algorithms to learning simple behaviors such as rocket jumping and dodging rockets. The next chapter discusses a popular use of genetic algorithms, which are combined with a rule-like representation.
|