5 ms·
Genetic algorithms are in a certain sense doing things in the stupidest possible way. Whenever it creates a random mutation, A better algorithm programed with
by indrax 17y ago
Genetic algorithms are in a certain sense doing things in the stupidest possible way.
Whenever it creates a random mutation, A better algorithm programed with at least a little knowledge of the problem could make a change that is more likely to be beneficial. An educated guess instead of an uneducated one.
Since we know a fair bit about image compression, regular algorithms just take their own best guess without any selection step. (or an I wrong about this?)
Unless you really don't know anything about how to solve the problem, you should be able to make an algorithm thats faster by mutating the population more intelligently.
- roundsquare 17y agoIts a fair point, but genetic algorithms could be used on known problems to discover something new. You might think you know enough to try an educated guess, but there might be some area of the search space that the theory you are using hasn't covered. That being said, I don't know anything about image compression and we may well have a provably best way to do this...
- indrax 17y agoI've thought about your comment for a while. If you want to test for potentially interesting search space, you should do so systematically, not randomly. A random scattering could leave vast areas unsearched, and you might not even know. (I think you'd have to look at a visualization that happened to make it clear, or do some other analysis.) Covering the entire search space evenly is a better approach to start with. Increasing the frequency around high scoring points would automatically dig down into interesting regions.