What sort of problems were those? Have they ever been succesful on a task like "give us a good vm program"?
Generally they don't get given tasks like that, no. Usually it is more specific. I tried to find some cases where GP has been more successful than humans, but it's a bit hard to dig a vague reference like that out quickly (though I assure you they're out there). Try looking at the work of Angeline and maybe Banzhaf or Nording. 'Writing a vm program' is still beyond the capabilities of GP, but they are getting more sophisticated all the time.
GP is not a panacea. Obviously NFL does apply to GPs and they are better at some problems than others, though that didn't seem to be the way you were referencing NFL in your 1st post.
I assume careful construction of representation and mapping is required to get a useful result
Well not really - you wouldn't be evolving a representation, you'd be evolving kernel code. The atomic functions you give to a GP run matter, but not as much as you might think.
I wasn't referring to planning at a meta level (eg.where is linux going?) Rather, I was thinking about planning structures and interfaces
Well if you look at my original post I didn't suggest trying to evolve a complete kernel through GP, just small chunks of it.
Evolving an algorithm is easy, but interoperation of different algorithms are hard, because it is hard to find an objective function that can differentiate between good and bad interfaces.
Depends on what you mean by good and bad interfaces. I don't suppose GP will ever be able to build a kernel from the ground up on its own. (Although I'd be interested to see what the results would be from gradually allowing a GP more and more control to find out where it became unfeasible.) A lot of work has been done on program structure, and GP is quite capable of producing subroutines and control structures etc. You said in your original post that it wouldn't be possible to evolve code more complex than simple functions which isn't correct in general.
I think the biggest difficulty would actually be building a test environment for the GP and ensuring it could not overfit the test data.
Even then, it won't better than human designed code
Err... why? GP has produced lots of impressive solutions to problems that consist of better code than produced by any human before.
Start with bad mapping (such as genotype consists of ascii characters and phenotype is a copy of genotype until first EOF) and you wouldn't even be able to get any compilable code in a lifetime
You've obviously ever even looked at genetic programming (as opposed to genetic algorithms) before. For a start, there is no mapping (genotype == phenotype), and all code produced is syntactically valid.
GAs are not good at "planning" kind of tasks.
Linus's point was that there wasn't much planning involved! I was working from that assumption, though I'm not sure to what degree I agree with him.
You should check "no free lunch theorem."
Thanks, I'm well aware of the NFL theorems. Perhaps you'd like to say how it applies in this case?
And don't EVER make the mistake that you can design something better than what you get from ruthless massively parallel trial-and-error with a feedback cycle. That's giving your intelligence _much_ too much credit.
Hmm... anyone ever tried to build an O/S with genetic programming? How about using GP to modify little chunks of the Linux kernel? I'm guessing the problem would be testing thousands of kernels...
I was thinking of doing exactly the same thing - I even registered a suitable domain name about a year ago - but I didn't have time to develop it. Anyway, Amazon's list covers most things.
>He grows up, travels off to an engineering
>school somewhere
Travels to an engineering school? Where? Afghanistan? There's almost no infrastructure left; currently we're bombing rubble and shacks. The West? With what money?
I've always assumed it was the other way around - people with terrible handwriting start using computers just so they can use a keyboard rather than a pen. I don't think I've ever met a proper computer geek who can write legibly.
I was quite disappointed to see that they've left the Celts, my usual choice of civ, out of the game this time. Any idea how they decide on what civs to include/leave out?
And because no organisms use lasers as a defensive weapon, they cannot exist?
evolution dictates that any lifeform that does not take full advantage of its enviroment will not survive to procreate
Evolution dictates nothing of the sort. All that happens is the fittest of any set of organisms is more likely to survive. There is no guarantee of optimality (otherwise I would currently be making a fortune with my perfect genetic algorithm-evolved stock market model).
That's not what they're doing. You're listening to Taliban's propoganda. The US is bombing Taliban targets in Afghanistan
Yes that's right. The
BBC and the Red Cross are just Taliban mouthpieces. To bomb Red Cross installations once seems like carelessness; to do it twice must be the most amazing piece of bastard stupidity I have seen. Remember, the US army were given the locations of all the Red Cross buildings in Afghanistan.
You couldn't make it up. All across America, people are commendably donating cash to the Red Cross. The Red Cross then use the cash to transport desperately-needed food to the Afganistan population. At which point, it is bombed out of existence by US military forces. You don't have to be an anti-US pacifist to see that something is seriously wrong there.
To the best of my knowledge, this wasn't even reported in the US (at least, I couldn't find it on CNN.com). Was it free speech you were fighting for again?
Unfortunately, no amount of understanding of *why* on our part will ever convince the people who destroyed the WTC to stop doing it. They don't care if we understand--they want us destroyed.
Possibly true, but surely it's worth a try anyway. Understanding other points of view is never a bad thing. Everyone does things for a reason, real or imagined.
Yeah, great idea. Perhaps you could apply the same logic to the criminal justice system as well.
Convicted of stealing? Cut their families' hands off!
Convicted of murder? Shoot their families and friends!
Rape? Castrate all their male acquantainces.
Terrorism? Bomb their women and children!
Maybe it would cut down crime... or maybe it would lead to the overthrow of the government by people sickened at the sight of burning babies.
Meaning, the keys aren't labeled, so re-mapping the character set doesn't result in a conflict with the labels
But that makes no sense. I mean, semantically it does, but it can hardly be described as a feature. By that rationale the keys on a standard keyboard should be blank so it doesn't matter what layout you might like to use.
Why not make the pyramid transparent and label the keys?
I wonder if there's an official definition of a currency.
Ability to earn cash in Everquest... check.
Ability to buy things in EQ... check.
Exchange rate... check, if unofficially.
Actually I'd be interested to see a graph of the exchange rate of EQ cash v dollars - which is the better long-term investment? With the impending recession it might actually be a good, (if somewhat risky!) investment to transfer your savings to EQ.
Seriously, those interdimensional regulations are for your own safety. You really don't want bullsquids dropping into your office, even if you've installed mirrors on your monitor.
IF you look at biological virology, and compare it to computer viruses, the similarities are striking.
I'm waiting for the first worm to appear that has a quasi-genetic structure.
Create a population of worms, and give each worm a few chromosomes, and some code that allows it to propagate using strategies determined by its genetic material. Give the worms an initial state that allow it to exploit some basic M$ vulnerabilities, and release a few hundred.
Every time a worm infects a new system, it looks for any other genetic-based worms. They've also been successful in infecting the system, so get the worms to mate and produce a new generation of a few tens of individuals from their genes (plus a few modifications).
Rinse and repeat.
Re:First Application Of This...
on
Share The Pi!
·
· Score: 1
I suspect this would be slightly impractical. Assuming the Metallica MP3 takes up 3Mb... this is ~24 million bits.
The chances of any 24 million consecutive random bits being the same as the bits of this MP3 is 0.5 ^ 24,000,000.
This is a rather small number. For any digit of Pi, the chances of it being the start of a sequence that encodes the MP3 is a half times itself 24 million times.
I think any description of the computing effort required would involve phrases like "current age of the universe" and "quantum computers the size of the solar system". I suspect your distributed computing effort would be looking for a long time.
Regional DVD coding was devised to protect the theatrical distribution market for motion pictures in international markets. It is simply impossible with present technologies to supply film prints of a movie to all of the theaters around the world at the same time. Motion pictures released by the major studios are generally released first in the Untied States and subsequently overseas. For this reason, motion pictures are released to theaters in countries in a "staggered" sequence. After the theatrical exhibition of a motion picture in a particular country, it is then released to the pay-per-view, video and television markets. DVDs are regionally coded to prevent them from being imported into countries where the motion picture has not yet completed its theatrical release. Without such protections, motion picture theatrical distributors and exhibitors abroad could lose a significant portion of their audiences to advance DVD viewing. The lost theatrical revenues could result in theater closures, lost jobs, depriving consumers throughout the world from seeing motion pictures on the big screen. A similar impact has occurred in some worldwide markets where illegal imports are unchecked.
... but why the bastards actually do stagger the releases is another question I'd like answered. "Piracy" would surely decrease in other countries if the films were released simultaneously. If the only copies available are illegal ones, then people are going to watch it (the releases in the UK can be 3 or more months after the US release).
I haven't read the book , but I'd like to see less reliance on file hierarchies. They might be a useful categorization at the system level, but I think they should be invisible to the general user.
Why should I need to tell an application to install itself to c:\program files\company\ellipsis or that its config file is at ~/.thingy ? All that information should be hidden from me and handled by the O/S. I should only need to know about the hierarchy structure if I'm actually writing code.
I'm not sure how you'd organise your documents though. Still in a hierarchy, but a separate one? (at least, separate from the point of view of the user). Perhaps you could retrieve them intelligently through the application you use them with (although its difficult to see how that might work).
Slashdot Mirror
Backslashdot?
or (4), given the current state of the world economies, all the coders are trying to keep a bit of job security.
OK, try this link, a paper by John Koza:
p df
http://www.genetic-programming.com/gpemcontrol.
GPs finding patentable inventions.
What sort of problems were those? Have they ever been succesful on a task like "give us a good vm program"?
.where is linux going?) Rather, I was thinking about planning structures and interfaces
Generally they don't get given tasks like that, no. Usually it is more specific. I tried to find some cases where GP has been more successful than humans, but it's a bit hard to dig a vague reference like that out quickly (though I assure you they're out there). Try looking at the work of Angeline and maybe Banzhaf or Nording. 'Writing a vm program' is still beyond the capabilities of GP, but they are getting more sophisticated all the time.
GP is not a panacea. Obviously NFL does apply to GPs and they are better at some problems than others, though that didn't seem to be the way you were referencing NFL in your 1st post.
I assume careful construction of representation and mapping is required to get a useful result
Well not really - you wouldn't be evolving a representation, you'd be evolving kernel code. The atomic functions you give to a GP run matter, but not as much as you might think.
I wasn't referring to planning at a meta level (eg
Well if you look at my original post I didn't suggest trying to evolve a complete kernel through GP, just small chunks of it.
Evolving an algorithm is easy, but interoperation of different algorithms are hard, because it is hard to find an objective function that can differentiate between good and bad interfaces.
Depends on what you mean by good and bad interfaces. I don't suppose GP will ever be able to build a kernel from the ground up on its own. (Although I'd be interested to see what the results would be from gradually allowing a GP more and more control to find out where it became unfeasible.) A lot of work has been done on program structure, and GP is quite capable of producing subroutines and control structures etc. You said in your original post that it wouldn't be possible to evolve code more complex than simple functions which isn't correct in general.
I think the biggest difficulty would actually be building a test environment for the GP and ensuring it could not overfit the test data.
Even then, it won't better than human designed code
Err... why? GP has produced lots of impressive solutions to problems that consist of better code than produced by any human before.
Start with bad mapping (such as genotype consists of ascii characters and phenotype is a copy of genotype until first EOF) and you wouldn't even be able to get any compilable code in a lifetime
You've obviously ever even looked at genetic programming (as opposed to genetic algorithms) before. For a start, there is no mapping (genotype == phenotype), and all code produced is syntactically valid.
GAs are not good at "planning" kind of tasks.
Linus's point was that there wasn't much planning involved! I was working from that assumption, though I'm not sure to what degree I agree with him.
You should check "no free lunch theorem."
Thanks, I'm well aware of the NFL theorems. Perhaps you'd like to say how it applies in this case?
And don't EVER make the mistake that you can design something better than what you get from ruthless massively parallel trial-and-error with a feedback cycle. That's giving your intelligence _much_ too much credit.
Hmm... anyone ever tried to build an O/S with genetic programming? How about using GP to modify little chunks of the Linux kernel? I'm guessing the problem would be testing thousands of kernels...
Sorry, I don't understand. If Slashdot wants to interview me, then no problem, I don't mind typing in a few answers...
This was a bit of a soulless interview. Couldn't we just have waited until he was a bit less busy?
I was thinking of doing exactly the same thing - I even registered a suitable domain name about a year ago - but I didn't have time to develop it. Anyway, Amazon's list covers most things.
>He grows up, travels off to an engineering
>school somewhere
Travels to an engineering school? Where? Afghanistan? There's almost no infrastructure left; currently we're bombing rubble and shacks. The West? With what money?
I've always assumed it was the other way around - people with terrible handwriting start using computers just so they can use a keyboard rather than a pen. I don't think I've ever met a proper computer geek who can write legibly.
I was quite disappointed to see that they've left the Celts, my usual choice of civ, out of the game this time. Any idea how they decide on what civs to include/leave out?
And because no organisms use lasers as a defensive weapon, they cannot exist?
evolution dictates that any lifeform that does not take full advantage of its enviroment will not survive to procreate
Evolution dictates nothing of the sort. All that happens is the fittest of any set of organisms is more likely to survive. There is no guarantee of optimality (otherwise I would currently be making a fortune with my perfect genetic algorithm-evolved stock market model).
That's not what they're doing. You're listening to Taliban's propoganda. The US is bombing Taliban targets in Afghanistan
Yes that's right. The
BBC and the Red Cross are just Taliban mouthpieces. To bomb Red Cross installations once seems like carelessness; to do it twice must be the most amazing piece of bastard stupidity I have seen. Remember, the US army were given the locations of all the Red Cross buildings in Afghanistan.
You couldn't make it up. All across America, people are commendably donating cash to the Red Cross. The Red Cross then use the cash to transport desperately-needed food to the Afganistan population. At which point, it is bombed out of existence by US military forces. You don't have to be an anti-US pacifist to see that something is seriously wrong there.
To the best of my knowledge, this wasn't even reported in the US (at least, I couldn't find it on CNN.com). Was it free speech you were fighting for again?
Unfortunately, no amount of understanding of *why* on our part will ever convince the people who destroyed the WTC to stop doing it. They don't care if we understand--they want us destroyed.
Possibly true, but surely it's worth a try anyway. Understanding other points of view is never a bad thing. Everyone does things for a reason, real or imagined.
Yeah, great idea. Perhaps you could apply the same logic to the criminal justice system as well.
Convicted of stealing? Cut their families' hands off!
Convicted of murder? Shoot their families and friends!
Rape? Castrate all their male acquantainces.
Terrorism? Bomb their women and children!
Maybe it would cut down crime... or maybe it would lead to the overthrow of the government by people sickened at the sight of burning babies.
Meaning, the keys aren't labeled, so re-mapping the character set doesn't result in a conflict with the labels
But that makes no sense. I mean, semantically it does, but it can hardly be described as a feature. By that rationale the keys on a standard keyboard should be blank so it doesn't matter what layout you might like to use.
Why not make the pyramid transparent and label the keys?
I wonder if there's an official definition of a currency.
Ability to earn cash in Everquest... check.
Ability to buy things in EQ... check.
Exchange rate... check, if unofficially.
Actually I'd be interested to see a graph of the exchange rate of EQ cash v dollars - which is the better long-term investment? With the impending recession it might actually be a good, (if somewhat risky!) investment to transfer your savings to EQ.
Seriously, those interdimensional regulations are for your own safety. You really don't want bullsquids dropping into your office, even if you've installed mirrors on your monitor.
IF you look at biological virology, and compare it to computer viruses, the similarities are striking.
I'm waiting for the first worm to appear that has a quasi-genetic structure.
Create a population of worms, and give each worm a few chromosomes, and some code that allows it to propagate using strategies determined by its genetic material. Give the worms an initial state that allow it to exploit some basic M$ vulnerabilities, and release a few hundred.
Every time a worm infects a new system, it looks for any other genetic-based worms. They've also been successful in infecting the system, so get the worms to mate and produce a new generation of a few tens of individuals from their genes (plus a few modifications).
Rinse and repeat.
I suspect this would be slightly impractical. Assuming the Metallica MP3 takes up 3Mb... this is ~24 million bits.
The chances of any 24 million consecutive random bits being the same as the bits of this MP3 is 0.5 ^ 24,000,000.
This is a rather small number. For any digit of Pi, the chances of it being the start of a sequence that encodes the MP3 is a half times itself 24 million times.
I think any description of the computing effort required would involve phrases like "current age of the universe" and "quantum computers the size of the solar system". I suspect your distributed computing effort would be looking for a long time.
... but why the bastards actually do stagger the releases is another question I'd like answered. "Piracy" would surely decrease in other countries if the films were released simultaneously. If the only copies available are illegal ones, then people are going to watch it (the releases in the UK can be 3 or more months after the US release).
Hmm. That's what I do all day anyway.
I haven't read the book , but I'd like to see less reliance on file hierarchies. They might be a useful categorization at the system level, but I think they should be invisible to the general user.
Why should I need to tell an application to install itself to c:\program files\company\ellipsis or that its config file is at ~/.thingy ? All that information should be hidden from me and handled by the O/S. I should only need to know about the hierarchy structure if I'm actually writing code.
I'm not sure how you'd organise your documents though. Still in a hierarchy, but a separate one? (at least, separate from the point of view of the user). Perhaps you could retrieve them intelligently through the application you use them with (although its difficult to see how that might work).
>Comments, Suggestions?
:)
I think it was a well argued comment. I'm going to copy it for posterity
---