The evolutionary algorithm is not the only way [...] It’s the best cheater I know, but likely not the only one.
Unless you meant to imply a specific problem (and very probably even then), evolutionary algorithms are actually pretty stupid. I’ll even go on a limb and claim that the evolutionary algorithm is the smartest of the stupid algorithms, where “stupid” means approximately “I understand nothing about the problem except I can tell if some solutions are better than others, if I’m given examples”.
Of course, if the problem is complicated enough that might be the best we can do.
But the main point of my article is this innovating aspect of the Evolutionary Algorithm. Unforeseen solutions delivered,
I’m not sure what you mean by “innovating”. A solution I receive from any algorithm that searches (rather than verifies) solutions will usually be unforseen. (If I foresaw it, I wouldn’t need to search for it, I’d just test it.)
Why not? If I used an EA to get it, that basically means “I don’t know how to solve the problem, so I’ll just use the best method I know of for trying random solutions”.
Also, I’m the world record holder at looking like myself, that doesn’t mean that I’m smarter than anyone else, particularly in the sense of knowing how to build a person that looks like myself.
“I don’t know how to solve the problem, so I’ll just use the best method I know of for trying random solutions”
If your random guessing will provide a (previously unknown) solution, very well. But it probably won’t.
I am not talking about “maybe it would”, I talk about “it did, indeed”.
I’m the world record holder at looking like myself
Everybody has few such records, but those are worthless. Some people however, solved a difficult puzzle. On this particular site I gave you a link, such a competition is going on. Maybe it reminds someone to the LW’s PD agents competition?
Anyway, I don’t hold any record there. An algorithm I designed and called it Pack’n’tile holds some. Follow the links, download the program and try it, if you want.
Sorry, I was unclear. By “best method I know for trying random solutions” I meant evolutionary algorithms. (Which I think of as “guess randomly, then mix guesses randomly, then mutate guesses randomly, then select randomly biased towards the the best you found, then repeat from step two”. Of course, there’s a bit of smartness needed when applying the randomness, but still.)
Some people however, solved a difficult puzzle.
I think we’re having mostly a terminology disagreement. I tend to think of EA as “finding a solution” rather than “solving the problem”, which I agree is not the most logical and precise use of language.
On another subject, I fear I may have offended you. If so, I apologize, and kudos for keeping calm enough to make it hard for me to be sure :)
I specifically said that the algorithms are stupid. That wasn’t meant to disparage anyone that uses them. I well know that it’s not at all trivial to write such an algorithm, and that there are good and bad ways of doing it, and that one can put a lot of cleverness in one. The authors of an algorithms that “won” a record in an important problem are very probably very smart people. But the algorithm itself may still be stupid, in the sense that it’s closer to brute force than actually finding the solution with a minimum of (computing) effort.
Technically speaking EA is stupid in the sense it’s very brief. The actual implementation is another matter.
But what is important in this context is the following: The algorithm’s results in this matching context are quite sloppy in the sense, that the squares don’t even touch each other to gain some more space. Still,the whole circle setting is so clever, that it can afford this generosity and still wins! After then, some humans often polish the evolved solution and claim the victory. What’s perfectly fine, the whole log exists.
Just curious, as I’m not familiar with that particular problem: are any of those records on “density of packing per FLOP”, or just “density of packing”?
Unless you meant to imply a specific problem (and very probably even then), evolutionary algorithms are actually pretty stupid. I’ll even go on a limb and claim that the evolutionary algorithm is the smartest of the stupid algorithms, where “stupid” means approximately “I understand nothing about the problem except I can tell if some solutions are better than others, if I’m given examples”.
Of course, if the problem is complicated enough that might be the best we can do.
I’m not sure what you mean by “innovating”. A solution I receive from any algorithm that searches (rather than verifies) solutions will usually be unforseen. (If I foresaw it, I wouldn’t need to search for it, I’d just test it.)
They hold some world records on the density of packing.
If you held just one, would you call yourself stupid?
I guess not.
Why not? If I used an EA to get it, that basically means “I don’t know how to solve the problem, so I’ll just use the best method I know of for trying random solutions”.
Also, I’m the world record holder at looking like myself, that doesn’t mean that I’m smarter than anyone else, particularly in the sense of knowing how to build a person that looks like myself.
If your random guessing will provide a (previously unknown) solution, very well. But it probably won’t.
I am not talking about “maybe it would”, I talk about “it did, indeed”.
Everybody has few such records, but those are worthless. Some people however, solved a difficult puzzle. On this particular site I gave you a link, such a competition is going on. Maybe it reminds someone to the LW’s PD agents competition?
Anyway, I don’t hold any record there. An algorithm I designed and called it Pack’n’tile holds some. Follow the links, download the program and try it, if you want.
Sorry, I was unclear. By “best method I know for trying random solutions” I meant evolutionary algorithms. (Which I think of as “guess randomly, then mix guesses randomly, then mutate guesses randomly, then select randomly biased towards the the best you found, then repeat from step two”. Of course, there’s a bit of smartness needed when applying the randomness, but still.)
I think we’re having mostly a terminology disagreement. I tend to think of EA as “finding a solution” rather than “solving the problem”, which I agree is not the most logical and precise use of language.
On another subject, I fear I may have offended you. If so, I apologize, and kudos for keeping calm enough to make it hard for me to be sure :)
I specifically said that the algorithms are stupid. That wasn’t meant to disparage anyone that uses them. I well know that it’s not at all trivial to write such an algorithm, and that there are good and bad ways of doing it, and that one can put a lot of cleverness in one. The authors of an algorithms that “won” a record in an important problem are very probably very smart people. But the algorithm itself may still be stupid, in the sense that it’s closer to brute force than actually finding the solution with a minimum of (computing) effort.
Technically speaking EA is stupid in the sense it’s very brief. The actual implementation is another matter.
But what is important in this context is the following: The algorithm’s results in this matching context are quite sloppy in the sense, that the squares don’t even touch each other to gain some more space. Still,the whole circle setting is so clever, that it can afford this generosity and still wins! After then, some humans often polish the evolved solution and claim the victory. What’s perfectly fine, the whole log exists.
Just curious, as I’m not familiar with that particular problem: are any of those records on “density of packing per FLOP”, or just “density of packing”?
What is that all about, you can best see here
OK, thanks, I’ll look if I have time, that’s a bit too much info to go through right now.