Genetic Algorithms in 400 Words or Less

image: Fox

I’m a Genetic Algorithm fanboy. Maybe it’s Hollywood’s fault. They’ve spent billions of dollars on CGI and Shia LeBoeuf only to horrify and deter Americans away from autonomous technology forever. This only fuels my interest.

I think the biggest hurdle for most GA newbies is comprehending of the concept. We’re breeding mathematics here people. Getting beyond that opens up a world of creepy potential.

The nutshell:

Imagine a snowy tundra. We’d like to find a very good species of rabbit to live in our tundra. So we borrow our neighbor’s Scion xB and stuff it full of rabbits. Rabbit species can be distinguished via a handful of characteristics, like its fur color, fur thickness, weight, etc. Each rabbit in the Scion is quite different, such that the herd represents a fairly good cross-section of all rabbit species.

So now you release the Scion xB full of bunnies onto the tundra and let nature take its course. After a generation passes, the thin-furred rabbits have perished, unable to withstand the cold. By the 3rd generation, the dark-colored rabbits have disappeared, eaten by predators. By the 5th generation, the larger rabbits have died, unable to find enough food to support their size. Eventually we are left with a thriving species of white, thick-furred, light-weight rabbits.

+2 Darwin Points.

So to apply this to a search algorithm, you only need to do two things: define your system, and define your creature. Your system should be a broad, testable entity (for example, “a horse race”) with a clear rankable fitness (“money”), while your creature (“a betting system”) must contain easily adjustable parameters.

With proper mutation, cross-over and parent selection, over time you may approach a near-optimal design.

The creepiness of these algorithms is entirely lost in any description and must be experienced to fully appreciate. Check out my favorite breveCreatures screensaver, a very visual and entertaining demonstration of the genetic algorithm scope.

But as always, be careful.