GAs are Good For Uninformed Search

• If you've got several, even dozens, of parameters and you have no idea how to optimize, AI people often use a genetic algorithm.
• Encode each parameter as a unit, and the full parameter set is a gene.
• Run the system with the parameters, and see how well it does.
• What's the evaluation function?
• Use good parameters sets to create new parameter sets.
• This can be combined with local hill climbing.
• It is important that related parameters travel around together.