Re:Top 15 games as posted by 1up:
on
20 Years of NES
·
· Score: 1
it wasn't chosen by the authors, it was voted on by the people.
A lot of people owned Zelda II, therefore it got a lot of votes. Same reason why Duck Hunt is on the list.
It's one of the 15 "hidden gems" mentioned on the site. (The writers picked much more consistently good games than the site's voters did, but I guess that's not really surprising -- the votes naturally gravitate toward games that a lot of people owned, like Zelda II, even when the quality was only so-so.)
Two of my favorite NES games were by Tengen. Gauntlet was huge, challenging, and quite well-designed; to this day I've never made it past the dragon on the last level. They also made a version of Tetris that was really embarrassingly superior to Nintendo's own. The two-player game was especially good.
See, the problem with the distinction you make is that we still don't know what the Mechanism (with a capital 'M') behind gravity is... What you're calling a mechanism is actually a description of the behavior.
I know, but what we know doesn't really matter. Newton could reasonably have hoped to discover something that looked like a mechanism for gravity, say, invisible strings, or angels shoving things. A descriptive equation perhaps can be an end point, but there's no logical requirement for it to be, right? You can write down an equation for how much hydrogen combines with 1g oxygen to make water with nothing left over, but it has a funny constant, and there turns out to be a complicated system underneath, which is better thought of in terms of atoms and molecules -- real objects -- than as a vast descriptive equation. Gravity could have been like that (and, hey, might yet be).
You can't hope for any better, because observation, interpretation, and description are the only valid modalities of science.
Maybe in an extreme theoretical sense, sure. In practice I think it normally does make sense to use science to talk about causation and mechanisms. That is, it's more useful to understand genetic recombination as the most likely known mechanism for observed patterns of inheritance and for many other things, rather than as the best known description of patterns-of-inheritance-plus-thousands-of-other-re levant-observations all taken together.
I think the parent poster was implying a distinction between description and mechanism, although he didn't put it very clearly. Newton's equations describe the workings of gravity, but they're independent of what one might call mechanism; he did not discover whether the motion of bodies toward one another is due to curvature in the universe, or inverted ethereal rubber bands connecting all mass, or what. Mendel did something similar, formulating a rather precise description of how inheritance worked before the concept of a gene even existed.
However that workload is only 96 page loads per second on average. Even accounting for load nonuniformity, the original system would still have been total overkill.
I think you might be underestimating how non-uniform traffic could get, and how important it might have been to be ready for it from the start. They probably anticipated a large spike at the beginning from advertising (25x doesn't seem crazy), and felt that if customers couldn't get through the first time they tried, many of them would never come back - upgrading later would be too little, too late.
Re:Dumbest thing I've read all week...
on
The Evil in E-Mail
·
· Score: 1
I'm pretty sure it's standard to judge performance by splitting your positive and negative examples into a "training" set and a "test" set (preferably several times), no? I think that's the usual way to judge what's working and what needs tweaking. The numbers will be a bit inflated if your test set is too closely related to the training set, but at worst you've got an easy-to-report upper bound. I guess our difference here is that I think that if Skillicorn had any kind of promising results (as opposed to just an idea that didn't work well when he tested it on Enron data) then he would have managed to allude to the fact somehow; whereas you seem to be suggesting that he just doesn't want to say anything until somebody uses it successfully in a real-world application?
The original poster's point seems to be that the message features cited in the article are highly imprecise, which is clearly true, and a valid criticism of a classifier based on those features. Of course, the features might be better than they look, or they might be statistically useful in conjunction with other features; but then again they might not. I still don't see anything wrong with pointing out the problems.
Cobbling together statistics from word frequencies doesn't rely on knowing what the words mean. I don't see any reason for it not to work for any language. They'd just have to have a native speaker spend a bit of time coming up with a list of sensitive words like bomb.
Actually, you don't even have to do that -- the whole point of word frequency analysis is that sensitive words are detected automatically.
(You do need native speakers to help you with syntactic constructions and stuff if you want to do anything slightly subtle, of course.)
Re:Dumbest thing I've read all week...
on
The Evil in E-Mail
·
· Score: 1
As someone that has developed commercial systems that do latent semantic processing (and other sorts of text analysis), I'm soooooooo glad that you can tell if it works by a single article written for the layman.
He's basically right, though. If you have a technique that works, and you're explaining it for general audience, wouldn't you mention the technique's successes? If Skillicorn has any success stories, he's done a good job of hiding them.
It seems clear that all he's doing is applying standard document classification algorithms to the problem of classifying criminal vs. non-criminal. It's exactly the same thing as classifying documents into spam vs. non-spam, Shakespeare vs. non-Shakespeare, whatever. Presumably he's gathered a corpus of criminal and non-criminal e-mails and is in the process of tweaking the features potentially used by the algorithm in order to better distinguish the two groups. It sounds really dull, and evidently doesn't work very well yet.
Could it ever work? Hard to say. Without trying the method first, there's really no way to know whether word-level differences can predict criminality at all. So the attempt isn't without merit... but the grandparent's analysis of likely shortcomings is spot-on.
(As an aside, classifying the Unabomber's manifesto [compared to what?] sounds much likelier to work than analyzing brief e-mails sent by people who are trying to look innocent.)
Wikipedia is available over HTTP in a much more up-to-date, interactive and dynamic format than DVDs.
Unfortunately, in Wikipedia's case "dynamic" sometimes means repeated back-and-forth edits and disputes over an article. Presumably, the DVD will represent a somewhat stable and agreed-upon version, rather than just a snapshot at an arbitrary moment.
Last dump made: 2005-03-09 (30 days ago) Total size 50503MB (1460MB for just current revisions)
These are SQL dumps of the current and old article revision databases for each wiki. They can be read into a local database and directly used with the MediaWiki software (MySQL, PHP, Apache required).
These dumps are not suitable for viewing in a web browser or text editor unless you do a little preprocessing on them first.
Right, it's kind of like an implementation of bayesian spam filtering, but for other problem domains.
By Bayesian spam filtering, I think you mean general classification problems, in which case, yes, neural networks can implement classification - it's a stretch to say that McClelland and Rumelhart's did, because the possible output included most non-repeating combinations of English phonemes and is thus nearly infinite, but the principle is there.
Of course, you'd also need to correctly simulate *how* the brain is wired up to get any kind of beneficial processing.
I think you're overestimating the importance of processing speed and underestimating the importance of the above.
For starters, a whole lot of parsing of the input has to go on -- retinal images parsed into people and objects, sound streams broken up by source, language identified, relevant phonological boundaries determined, speech separated into words. Then you have to know what your goal is, and approach it at the right time: learning syntax or social conventions won't help a baby who doesn't know words or faces yet. And what's the output of "learning syntax", anyway? A list of rules? A network that can turn... something... into a sentence?
Throwing more nodes into a network doesn't get anywhere with these problems, whether the network "grows itself" or whether the programmer does the work. The big problem is structuring inputs and outputs to be complete, sensible, and not wrongly redundant, and perhaps arranging networks in sequences or graphs to separate information and model psychological findings of dissociations between tasks.
Also, there's a great deal of parallel processing in play. Your 100 THz processor can perform a zillion operations per second in order, which gives it far more flexibility than the brain has. Between neuron firing rate and communication time, I think (can't currently find the reference) the brain is limited to about 100 sequential operations per second.
That's astoundingly few. You can come up with a good chunk of a sentence in a second, and recognize a blurry familiar face in less. Parallel or not, I have difficulty imagining how one does that in so few chunks of time.
I doubt it would be too difficult to code -- if we knew the mechanism by which it proceded.
Its hard to code a procedure to replicate the working of the mind...if you don't know how the mind does it in the first place.
On the other hand, it might be that the reason we don't understand how the mind does certain things is that they're actually extremely complicated, and don't reduce very well to a programmable step-by-step algorithm nor to a simple and general mathematical learning structure. It's hard to tell, although I think it's telling that after decades of work, neither psychologists nor computer scientists can understand or replicate much of what babies do.
Sometimes the best way for a computer to learn something may not be the way a baby does it, anyway; c.f. chess.
Until I see this new process in the works, however, there is nothing that will make me believe it's better than finding another human who can *understand* what you are saying and the context to which you are implying.
Heh. Then there is nothing that will make you believe, etc., etc.
Certainly you can't do good translation without understanding syntax (which influences meaning and underlies word order) and context (to disambiguate synonyms and phrases with multiple interpretations). Machines aren't especially good at either one yet; ergo, machine translation will continue to be pretty crappy for the foreseeable future.
Funny thing is, though, even a crappy translation turns out to be tremendously useful in most practical contexts, and worlds better than none at all; a simple word-for-word translation is typically hard to read but still conveys the proper gist. That's why I don't get excited about automatic translation "advances" these days: there are really two purposes for machine translation. One is figuring out what a piece of speech or text trying to say, and the current technology is usually good enough for that. The other is making a translation of sufficient quality to save a human translator some work, and I think that won't happen for quite a few years yet. Anything in between adds very little.
(By the way, everything in natural language processing these days uses corpus learning techniques. Now if an improved technology had been developed manually by bilingual programmers who pulled the design out of their collective hats, then that would be a man-bites-dog story!)
But most photographers are probably not working "in connection with advertisements or commentaries related to the distribution or display of" the Bean, which makes this passage irrelevant. (It sounds to me like a copyright exemption for art critics.)
Is there a particular paper or author you'd recommend? I don't have time to sift through thousands either, but I'm interested enough to read one or two, if you know of a good review or an experiment with particularly clear results...
It's certainly a good idea, though not a new one. Such a language is called an interlingua -- I'd advise you to search for references, except the word appears to have several more common meanings. I know people are working on it, but of course both the representational (what form does the language take?) and translational problems are extremely hard.
So what happens when you stop taking the course of antibiotics halfway through? Well, where you previously had a bacterial population consisting of some bacteria with weak resistance, some with moderate resistance, and some with strong resistance, now you only have the latter two categories. And these are going to continue breeding, and your immune system is going to spend its resources fighting them equally, without preference as to which is more or less antibiotic-resistant -- which means more of the bacteria with greater resistance will survive and grow. OTOH, if you'd finished the antibiotics, only the most resistant bacteria would be left, and your immune system could probably finish them off on its own.
That's a fair theory... but I've never seen any evidence that's it's actually correct. Your scenario is that after a partial course of antibiotics, the moderate-to-strong bacteria survive for a moderate period before being finished off by the immune system. But by the same token, it seems that after a full course of antibiotics, then only the strongest bacteria would survive (worse), but for a shorter time (better). Which scenario leads to more antibiotic resistance in practice? How do we know?
Your point about resistance requiring energy (which also makes sense but isn't self-evident either) leads to an argument that the latter case could be worse, because if both moderately resistant and strongly resistant bacteria are left post-antibiotics, then the ones with moderate resistance will have more energy for propagation... so it could be better to stop early in order to leave some non-resistant bacteria around.
But it does seem like an empirical question, right? If you know of relevant research I'd be interested to see it...
This is also a "phone survey". Asking someone whether or not they know the difference is not the same as them knowing the difference.
Exactly. The article says "of [subset of web searchers], not even half say they can always tell which [ads] are paid."
Can you always tell which ads are paid? With Google, sure you can. But what about other search engines? Are you sure they all even display a distinction? If you're not sure, or haven't thought about it, or if you accidentally clicked on a paid ad once, then maybe you figure you can't always tell, so you say no.
If they'd sat people down at computers and asked them to identify which ads are paid, it seems likely that the numbers would have been much higher.
Eh, I never much liked Weis and Hickman, except the first... oh, four or five books of the Death Gate cycle. Their Dragonlance stuff just seemed kind of blah, one thing happening after another; it's hard to make out a coherent story.
If you read Dragonlance, pick up Richard A. Knaak. He's actually quite good, what I've read of him (try The Legend of Huma).
I've never understood this particular drawing of Escher's. To me it just looks like the top of the waterfall is... the top, and none of the water looks like it's flowing up in that direction. The support pillars make the top of the drawing look higher, and the stepped bricks don't have that much effect because I see the whole channel as tilted.
Slashdot Mirror
it wasn't chosen by the authors, it was voted on by the people. A lot of people owned Zelda II, therefore it got a lot of votes. Same reason why Duck Hunt is on the list.
It's one of the 15 "hidden gems" mentioned on the site. (The writers picked much more consistently good games than the site's voters did, but I guess that's not really surprising -- the votes naturally gravitate toward games that a lot of people owned, like Zelda II, even when the quality was only so-so.)
I belive there were several Tengen games...
Two of my favorite NES games were by Tengen. Gauntlet was huge, challenging, and quite well-designed; to this day I've never made it past the dragon on the last level. They also made a version of Tetris that was really embarrassingly superior to Nintendo's own. The two-player game was especially good.
See, the problem with the distinction you make is that we still don't know what the Mechanism (with a capital 'M') behind gravity is... What you're calling a mechanism is actually a description of the behavior.
e levant-observations all taken together.
I know, but what we know doesn't really matter. Newton could reasonably have hoped to discover something that looked like a mechanism for gravity, say, invisible strings, or angels shoving things. A descriptive equation perhaps can be an end point, but there's no logical requirement for it to be, right? You can write down an equation for how much hydrogen combines with 1g oxygen to make water with nothing left over, but it has a funny constant, and there turns out to be a complicated system underneath, which is better thought of in terms of atoms and molecules -- real objects -- than as a vast descriptive equation. Gravity could have been like that (and, hey, might yet be).
You can't hope for any better, because observation, interpretation, and description are the only valid modalities of science.
Maybe in an extreme theoretical sense, sure. In practice I think it normally does make sense to use science to talk about causation and mechanisms. That is, it's more useful to understand genetic recombination as the most likely known mechanism for observed patterns of inheritance and for many other things, rather than as the best known description of patterns-of-inheritance-plus-thousands-of-other-r
I think the parent poster was implying a distinction between description and mechanism, although he didn't put it very clearly. Newton's equations describe the workings of gravity, but they're independent of what one might call mechanism; he did not discover whether the motion of bodies toward one another is due to curvature in the universe, or inverted ethereal rubber bands connecting all mass, or what. Mendel did something similar, formulating a rather precise description of how inheritance worked before the concept of a gene even existed.
However that workload is only 96 page loads per second on average. Even accounting for load nonuniformity, the original system would still have been total overkill.
I think you might be underestimating how non-uniform traffic could get, and how important it might have been to be ready for it from the start. They probably anticipated a large spike at the beginning from advertising (25x doesn't seem crazy), and felt that if customers couldn't get through the first time they tried, many of them would never come back - upgrading later would be too little, too late.
I'm pretty sure it's standard to judge performance by splitting your positive and negative examples into a "training" set and a "test" set (preferably several times), no? I think that's the usual way to judge what's working and what needs tweaking. The numbers will be a bit inflated if your test set is too closely related to the training set, but at worst you've got an easy-to-report upper bound. I guess our difference here is that I think that if Skillicorn had any kind of promising results (as opposed to just an idea that didn't work well when he tested it on Enron data) then he would have managed to allude to the fact somehow; whereas you seem to be suggesting that he just doesn't want to say anything until somebody uses it successfully in a real-world application?
The original poster's point seems to be that the message features cited in the article are highly imprecise, which is clearly true, and a valid criticism of a classifier based on those features. Of course, the features might be better than they look, or they might be statistically useful in conjunction with other features; but then again they might not. I still don't see anything wrong with pointing out the problems.
Cobbling together statistics from word frequencies doesn't rely on knowing what the words mean. I don't see any reason for it not to work for any language. They'd just have to have a native speaker spend a bit of time coming up with a list of sensitive words like bomb.
Actually, you don't even have to do that -- the whole point of word frequency analysis is that sensitive words are detected automatically.
(You do need native speakers to help you with syntactic constructions and stuff if you want to do anything slightly subtle, of course.)
As someone that has developed commercial systems that do latent semantic processing (and other sorts of text analysis), I'm soooooooo glad that you can tell if it works by a single article written for the layman.
He's basically right, though. If you have a technique that works, and you're explaining it for general audience, wouldn't you mention the technique's successes? If Skillicorn has any success stories, he's done a good job of hiding them.
It seems clear that all he's doing is applying standard document classification algorithms to the problem of classifying criminal vs. non-criminal. It's exactly the same thing as classifying documents into spam vs. non-spam, Shakespeare vs. non-Shakespeare, whatever. Presumably he's gathered a corpus of criminal and non-criminal e-mails and is in the process of tweaking the features potentially used by the algorithm in order to better distinguish the two groups. It sounds really dull, and evidently doesn't work very well yet.
Could it ever work? Hard to say. Without trying the method first, there's really no way to know whether word-level differences can predict criminality at all. So the attempt isn't without merit... but the grandparent's analysis of likely shortcomings is spot-on.
(As an aside, classifying the Unabomber's manifesto [compared to what?] sounds much likelier to work than analyzing brief e-mails sent by people who are trying to look innocent.)
Wikipedia is available over HTTP in a much more up-to-date, interactive and dynamic format than DVDs.
Unfortunately, in Wikipedia's case "dynamic" sometimes means repeated back-and-forth edits and disputes over an article. Presumably, the DVD will represent a somewhat stable and agreed-upon version, rather than just a snapshot at an arbitrary moment.
Let anyone submit a program that produces, with no inputs, one of the major natural language corpuses as output.
:)
Ooh! Finally, a use for my infinite number of monkeys.
Right, it's kind of like an implementation of bayesian spam filtering, but for other problem domains.
By Bayesian spam filtering, I think you mean general classification problems, in which case, yes, neural networks can implement classification - it's a stretch to say that McClelland and Rumelhart's did, because the possible output included most non-repeating combinations of English phonemes and is thus nearly infinite, but the principle is there.
Of course, you'd also need to correctly simulate *how* the brain is wired up to get any kind of beneficial processing.
I think you're overestimating the importance of processing speed and underestimating the importance of the above.
For starters, a whole lot of parsing of the input has to go on -- retinal images parsed into people and objects, sound streams broken up by source, language identified, relevant phonological boundaries determined, speech separated into words. Then you have to know what your goal is, and approach it at the right time: learning syntax or social conventions won't help a baby who doesn't know words or faces yet. And what's the output of "learning syntax", anyway? A list of rules? A network that can turn... something... into a sentence?
Throwing more nodes into a network doesn't get anywhere with these problems, whether the network "grows itself" or whether the programmer does the work. The big problem is structuring inputs and outputs to be complete, sensible, and not wrongly redundant, and perhaps arranging networks in sequences or graphs to separate information and model psychological findings of dissociations between tasks.
Also, there's a great deal of parallel processing in play. Your 100 THz processor can perform a zillion operations per second in order, which gives it far more flexibility than the brain has. Between neuron firing rate and communication time, I think (can't currently find the reference) the brain is limited to about 100 sequential operations per second.
That's astoundingly few. You can come up with a good chunk of a sentence in a second, and recognize a blurry familiar face in less. Parallel or not, I have difficulty imagining how one does that in so few chunks of time.
I doubt it would be too difficult to code -- if we knew the mechanism by which it proceded.
Its hard to code a procedure to replicate the working of the mind...if you don't know how the mind does it in the first place.
On the other hand, it might be that the reason we don't understand how the mind does certain things is that they're actually extremely complicated, and don't reduce very well to a programmable step-by-step algorithm nor to a simple and general mathematical learning structure. It's hard to tell, although I think it's telling that after decades of work, neither psychologists nor computer scientists can understand or replicate much of what babies do.
Sometimes the best way for a computer to learn something may not be the way a baby does it, anyway; c.f. chess.
Until I see this new process in the works, however, there is nothing that will make me believe it's better than finding another human who can *understand* what you are saying and the context to which you are implying.
Heh. Then there is nothing that will make you believe, etc., etc.
Certainly you can't do good translation without understanding syntax (which influences meaning and underlies word order) and context (to disambiguate synonyms and phrases with multiple interpretations). Machines aren't especially good at either one yet; ergo, machine translation will continue to be pretty crappy for the foreseeable future.
Funny thing is, though, even a crappy translation turns out to be tremendously useful in most practical contexts, and worlds better than none at all; a simple word-for-word translation is typically hard to read but still conveys the proper gist. That's why I don't get excited about automatic translation "advances" these days: there are really two purposes for machine translation. One is figuring out what a piece of speech or text trying to say, and the current technology is usually good enough for that. The other is making a translation of sufficient quality to save a human translator some work, and I think that won't happen for quite a few years yet. Anything in between adds very little.
(By the way, everything in natural language processing these days uses corpus learning techniques. Now if an improved technology had been developed manually by bilingual programmers who pulled the design out of their collective hats, then that would be a man-bites-dog story!)
But most photographers are probably not working "in connection with advertisements or commentaries related to the distribution or display of" the Bean, which makes this passage irrelevant. (It sounds to me like a copyright exemption for art critics.)
Thank you!!
Is there a particular paper or author you'd recommend? I don't have time to sift through thousands either, but I'm interested enough to read one or two, if you know of a good review or an experiment with particularly clear results...
Thanks,
Chris
It's certainly a good idea, though not a new one. Such a language is called an interlingua -- I'd advise you to search for references, except the word appears to have several more common meanings. I know people are working on it, but of course both the representational (what form does the language take?) and translational problems are extremely hard.
So what happens when you stop taking the course of antibiotics halfway through? Well, where you previously had a bacterial population consisting of some bacteria with weak resistance, some with moderate resistance, and some with strong resistance, now you only have the latter two categories. And these are going to continue breeding, and your immune system is going to spend its resources fighting them equally, without preference as to which is more or less antibiotic-resistant -- which means more of the bacteria with greater resistance will survive and grow. OTOH, if you'd finished the antibiotics, only the most resistant bacteria would be left, and your immune system could probably finish them off on its own.
That's a fair theory... but I've never seen any evidence that's it's actually correct. Your scenario is that after a partial course of antibiotics, the moderate-to-strong bacteria survive for a moderate period before being finished off by the immune system. But by the same token, it seems that after a full course of antibiotics, then only the strongest bacteria would survive (worse), but for a shorter time (better). Which scenario leads to more antibiotic resistance in practice? How do we know?
Your point about resistance requiring energy (which also makes sense but isn't self-evident either) leads to an argument that the latter case could be worse, because if both moderately resistant and strongly resistant bacteria are left post-antibiotics, then the ones with moderate resistance will have more energy for propagation... so it could be better to stop early in order to leave some non-resistant bacteria around.
But it does seem like an empirical question, right? If you know of relevant research I'd be interested to see it...
Good grief. I admire your manual dexterity, but the more common gestures are quite a bit simpler.
Cat: make a fist for the cat's head, extending first two fingers in "V" shape for the ears.
Tinfoil: all fingers straight, hand flat.
Microwave: make a fist, representing the box.
I'm sure if you practice you'll get it down...
This is also a "phone survey". Asking someone whether or not they know the difference is not the same as them knowing the difference.
Exactly. The article says "of [subset of web searchers], not even half say they can always tell which [ads] are paid."
Can you always tell which ads are paid? With Google, sure you can. But what about other search engines? Are you sure they all even display a distinction? If you're not sure, or haven't thought about it, or if you accidentally clicked on a paid ad once, then maybe you figure you can't always tell, so you say no.
If they'd sat people down at computers and asked them to identify which ads are paid, it seems likely that the numbers would have been much higher.
Why is that stupid? A Google search can actually involve less typing, particularly if Google is your homepage.
1) click in address bar, type in "www.amazon.com", press Enter
2) type in "amazon", click "I'm feeling lucky"
Maybe #2 is lazy, but it's not dumb.
Eh, I never much liked Weis and Hickman, except the first... oh, four or five books of the Death Gate cycle. Their Dragonlance stuff just seemed kind of blah, one thing happening after another; it's hard to make out a coherent story.
If you read Dragonlance, pick up Richard A. Knaak. He's actually quite good, what I've read of him (try The Legend of Huma).
I've never understood this particular drawing of Escher's. To me it just looks like the top of the waterfall is... the top, and none of the water looks like it's flowing up in that direction. The support pillars make the top of the drawing look higher, and the stepped bricks don't have that much effect because I see the whole channel as tilted.
Anyone have advice on how to "see" this illusion?