You mean Matlab isn't considered the best?
I kid, I kid.
Kid you may, but Mathematica is a computer algebra system, which means its good at manipulating symbolic mathematics. Matlab is primarily used for vector/matrix manipulation and is more engineering-oriented. I wish people would realize that in spite of the many commonalities (including the prefix "Mat"), they are different products with different uses and audiences.
That's why the concept of statistical significance exists.
Significance only makes sense when the underlying distribution is known, such as the random sequence I listed as an example. When you have no clue what the underlying distribution is, and can NOT safely assume near-normality because of the central limit theorem, all bets are off. I just don't buy that the distribution of terrorist attacks is normal or even near-normal, not without some hard evidence.
Sure, that's why you have test sets to determine if the models the system learns from the data are useful or not. I think it's safe to assume that the scientists working on this are familiar with the basics of learning theory and modeling.
Blindly applying learning theory to collected data often leads to models that are apparently highly predictive on your test/validation sets, but are absolute garbage in unseen scenarios. You're assuming that by learning a "model" that the distribution of data remains the same even when this system goes into deployment. While that might be nice for some kinds of data, how could you even begin justifying this for *terrorist attacks*??
The attacks are surprises and random, how are they going to try to extrapolate patterns with computers?
Even better -- if you look in television static long enough, you are going to find a pattern. Either they've found some hidden predictor of attacks, or maybe someone needs a course in basic Ramsey theory, which deals with conditions under which order (patterns) must occur even in random noise.
Consider this example (*not* meant as an analogy for the discrete math nazis): you have an infinite sequence of completely random letters over the alphabet. What is the probability of finding "abc" repeated 15 times with a gap of exactly 10 letters between successive repeats? If the stream is indeed completely random, then the probability is non-zero and you will EVENTUALLY (probably) see the "pattern".
retain the skills of expert operators, especially as they rotate out of the Air Force
This sounds like they're setting themselves up for one of the less predictable problems with the expert systems of the '80s: namely the fact that experts sometimes don't want to transfer their hard-acquired knowledge into a box designed to replace them. But given that this is the Air Force, "orders" might be the solution.
I'm a good geek of all arts. But when I try to dabble in graphical design, I always fail spectacularly.
Get someone with actual talent to do it.
Geek 2:
You want a Marketing Pro, who can deliver the rain, handling the "Vision", while you can concentrate on the implementation
And they're both absolutely right. I would suggest starting with a free template and modifying the CSS / graphics. That saves you the initial legwork of choosing a design layout, colors, etc. Here are some sites with some really awesome templates and liberal licensing (CC for most I think): Free CSS Templates.org and Open source web design.
PSO differs from some other types of hill climbing algorithms in that after each iteration, the population converges upon the current highest ranked individual. In essence, the population is acting like a 'swarm', by constantly moving towards the best known solution.
This is *exactly* the problem with this branch of computational intelligence, stuff that you see at any CI/AI conference. PSO is a minor variation of stochastic hill-climbing -- it's a friggin heuristic. There is no guarantee that it will perform better than any of the myriad other "biologically-inspired" heuristic out there, including ACO, GAs, ESs, etc. See the No free lunch theorem.
IMO (and as an earlier poster said I am an AI researcher, albeit not in CV), the whole field is trying to wean off a hangover from '89, when genetic algorithms blew people away with the fact that they were "biologically-inspired" and actually could do optimization. Now every minor heuristic variation of stochastic search has to have some cutesy Disney-like "biologically-inspired" back story -- why?? Just move on, damnit, or try to prove something about your heuristics (then again, see the NFL theorem).
That's all fine, but do we really another idiotic web 2.0 name for a startup? Vlingo?? REALLY!?!? Haven't we had enough of vongo, twitter, oyogi, flickr, xuqa, blinkx, sharkle, squidoo, zemq, diigo, frappr, joost, zingee, vyew, bebo?
Let's just wait for the official forensics rather than patched together rumours shall we?
Um...what are you doing on/. then? Seriously though, this is the place to come for some relatively informed speculation (see for example comments by the jet maintenance guy earlier in the thread).
Time itself does not slow down, time perception is subjective. Anyone who has ever ingested cannabis and looked at a ticking clock will tell you this. "From what people tell me", it is on the order of "experiencing" 3 seconds every second. No need for free-falling airplanes.
This is why the recent BitTorrent lawsuit against Comcast is so important...once they realize that they can't look inside encrypted packets, they're just going to block all p2p traffic. But even that is going to be hard, because at the encrypted UDP packet level, what really distinguishes a BT packet from, say, a Skype packet which is also encrypted by default? Screw encryption, what differentiates a DRM-free MP3 flying in from iTunes or Amazon from one coming through a modified BT protocol which uses port 80 and fake http headers?
In short, this is the dumbest idea and any implementation will be necessarily half-assed and is going to affect people.
SOAB! LiveJournal has been a great source for academic researchers in NLP, social networks and data mining, mainly because they don't mind you crawling their site and users have public "friend lists", etc. I doubt if that culture is going to be retained with a Russian owner. I was at a conference earlier this year, and met a guy from SixApart. He was talking about some sort of open API for crawling multiple social networking sites, but I guess that's probably going to be history now.
Not only that, but we now have some sort of government-manufactured rule-based system that assigns risk to 'potential terrorists'. Just wait for the inevitable leak of their methodology (via stolen laptops, incompetence, etc.) and you just gave real terrorists a way to evade suspicion. That's the problem with any "model" for suspicious behavior -- once its known, it's easily exploited.
All I really want for Christmas is an MP3 setup where several tech podcasts per day get downloaded automatically onto an MP3 player, and I carry it to my mini-van and listen to geek stuff on my hour-plus commute every day instead of mindless music.
It seems like you might want one of these. I believe the new zunes have a wireless dock that grabs podcast subscriptions from your pc while it charges. not sure if the new ipod touch does that, but i'd be surprised if it didn't.
So, how quickly would you run afoul of Intellectual Property laws doing this?
That's a great knee-jerk reaction. Without understanding the motivation behind the article, you assume that code scavenging means stealing other people's code. What they're really talking about is (legitimately) re-visiting code that you or other people have written, and then picking and modularizing bite-sized chunks. In other words, you would design a large program (mark I) and then go back and pick out useful parts, clean/debug them and have working modules (mark II) for the next project.
Also, for people who haven't read TFA, it's 9 short paragraphs long and barely an article. They talk about a "formal approach to code scavenging" without even coming close to explaining what exactly that MEANS.
Now imagine that the combination of an accelerometer and some clever programming of an off-center vibrate weight could simulate different weights in an object that doesn't actually change weight.
Just like you can fly by pulling on your shoelaces, nothing you do can change the weight, either real or "perceived", of an object sitting in your hand.
This is a pretty useless article. Doesn't really tell how he's planning on doing it.
.
I absolutely agree with you. Even the Computerworld (admittedly not the pinnacle of scientific reporting) article starts by saying "University of Ottawa student Kris Woodbeck is combining the neural processes we use to understand image data with the features of graphics processors." I don't even know where to begin with that statement. So he's come up with a model of neural image processing (a feat in itself)...and is mapping it to a GPU? This is like saying "we've figured out how to isolate stem cells from a source other than human embryos...and we used plastic petri dishes to do it!."
Second: 'The brain is very parallel. There's lots of things going on at once," he said. "Graphics processors are also very parallel'. OK, is this a science finding (i.e. a new image processing ALGORITHM based on the brain), or a systems paper (we came up with a parallel GPU version of an algorithm). I really hope he was misquoted, because otherwise it sounds like vaporware or untested hypotheses.
And then: 'For images, it might be when you took it, with what camera, with what exposure, that's about it. Then you're stuck with a red barn in rolling hills and I might know it was taken in California, but no one else does. How do you surface that metadata so it becomes much more searchable?' OK, now where did this come from again? neural processing? parallel GPU? and now inferring metadata?
Sorry, but getting a provisional patent is hardly a difficult thing (most universities in the US will file one for free if *YOU* think you might be on to something). Furthermore, all this would be more credible if they published some results and at least a brief description (which is allowable by patent law). Until I see some numbers, this is non-news.
That seems like an approach.
You ask 1,000 people.
Then you test them.
Now you have a "lying rate".
That'll only work if you don't tell them beforehand that they're going to be tested. If you did, only pathological liars would end up lying. And if you didn't, there's not a chance your study would be approved by any sane IRB (institutional review board for human subject trials). Or if you asked people if you could test them only after their response, what are the chances that the liars would take the test (and thus be revealed as "liars")? In all cases, your "lying rate" is very likely to be biased.
You just fired off a bunch of things that COULD be wrong with the studies. That's like me saying that your parents could have been the ones who assasinated JFK. Its not really based in fact
In science, if something *plausible* COULD be wrong with a study, it deserves to be analyzed before the study is assumed to be rigorous. This also applies to studies on the other side, i.e. the ones which claim marijuana cures death and stops global warming. In your analogy, you can't possibly give me any plausible evidence that my parents killed JFK. I pointed out a rather common methodological flaw (check the literature) with using self-reporting in smoked marijuana studies.
Look, I agree that people CAN cook cannabis and they SHOULD use a vaporizer if they want to smoke it. However plenty of people do smoke it in bowls. In fact I'd bet money that In the United States, most people smoke it without using a vaporizer. Even more people smoke it then cook it.
Would you say there's an honest culture of information about cannabis in the United States? I wouldn't. I don't know if you're being sarcastic or not, but I'd be all for a campaign to educate people on the safe use of marijuana. In the Netherlands, most coffee shops stock a vaporizer and a lot of Dutch people I've talked to would prefer to use a vaporizer. Ultimately, people will probably still smoke cannabis because of the social bonding aspect, but they should be educated about the alternatives. After that, it's a choice you make for yourself.
In case you were NOT being sarcastic, here are some websites that advocate safe marijuana use: safer choice, regulate, marijuana uses (not really an organization, but an emeritus harvard professor who's studying the positive uses of marijuana)
I'd like to address your point as well as that of one of my less coherent, ad hominem responders.
In your second point, you seem to be making the assumption that pot smokers who do not have emphysema will lie more often than those who do. While I am sure that we can imagine reasons why these rates would be different, do we have any evidence that they actually are? I would be interested to see if the researchers addressed this issue.
IANA social scientist, but this seems like its quite difficult to quantify. How can you tell when someone is lying if they say that they do not smoke marijuana? Short of testing their blood/hair/urine/etc or watching them continuously, there's no real objective way to tell if someone smokes cannabis or not. It also seems unlikely to me that someone who DOES agrees to undergo a drug test would lie about their drug use. Furthermore, someone who tests positive for cannabis might not actually SMOKE cannabis, as I mentioned. I also don't see how you could build redundancy into this type of question on a survey (perhaps a psychologist could correct me).
So we're left with only what people report. This is where I tend to turn to experience. In the United States, a lot of cannabis smokers I personally know would probably say "no" to the question on any type of questionnaire, even if it were anonymous. The point is that while you can't verify if someone is lying or not, I can think of a lot of plausible reasons why someone WOULD lie with draconian drug laws.
That said, it's of course still conjecture. That's one reason why I tend to distrust smoked marijuana studies -- they're usually targeting marijuana in its smoked form, not THC as a compound.
No worries, you will still get emphysema which is almost as unpleasant and tourchers you for longer. By the way, this explains why Cannabis has a ratio of emphysema to lung cancer that is quite different from cigarettes which is something that has been suspected for a while (hard to get good data because people generally smoke both).
This is righteous bullshit. Allow me to elaborate:
Cannabis studies have a history of being stooped in politics, alternating between pro and con. If you're talking about the ratio of emphysema to lung cancer, I'd be interested in how the population was sampled. Was it a random sampling of cannabis users? Was it people who showed up in a clinic with emphysema, and were then entered into the study? Consider the next point.
Suppose there are two equal-sized populations of cigarette smokers and cannabis smokers. Each population has exactly the same emphysema/lung cancer ratio. We'll assume that cannabis smokers don't smoke cigarettes for now, although if they did it would only make the study more dubious. If everyone reported accurately if there were a cannabis smoker or just cigarette smoker, we'd find approximately equal ratios. On the other hand, if some healthy cannabis smokers, out of fear of law enforcement or privacy reasons, reported themselves as cigarette smokers, the emphysema/lung cancer ratio in the sampled cannabis 'group' would appear to be much higher. I'd also doubt non-cannabis smokers reporting themselves as cannabis smokers.
The final point. You don't actually need to smoke cannabis, thereby removing all risk of emphysema and associated respiratory disorders. THC is fat-soluble, and so can be cooked or baked into anything that requires the use of fat or oil. Popular recipes include pasta and confectioneries, and I'm pretty sure none of those give you cancer. If you do choose to smoke cannabis, a vaporizer is often advocated in the Netherlands. I don't know what the health risks are, but they certainly seem to have a lot lower concentration of particulate matter (hence the name).
Slashdot Mirror
Kid you may, but Mathematica is a computer algebra system, which means its good at manipulating symbolic mathematics. Matlab is primarily used for vector/matrix manipulation and is more engineering-oriented. I wish people would realize that in spite of the many commonalities (including the prefix "Mat"), they are different products with different uses and audiences.
How about some neat animations as 256-byte executables -- 256b.com.
Significance only makes sense when the underlying distribution is known, such as the random sequence I listed as an example. When you have no clue what the underlying distribution is, and can NOT safely assume near-normality because of the central limit theorem, all bets are off. I just don't buy that the distribution of terrorist attacks is normal or even near-normal, not without some hard evidence.
Blindly applying learning theory to collected data often leads to models that are apparently highly predictive on your test/validation sets, but are absolute garbage in unseen scenarios. You're assuming that by learning a "model" that the distribution of data remains the same even when this system goes into deployment. While that might be nice for some kinds of data, how could you even begin justifying this for *terrorist attacks*??
Even better -- if you look in television static long enough, you are going to find a pattern. Either they've found some hidden predictor of attacks, or maybe someone needs a course in basic Ramsey theory, which deals with conditions under which order (patterns) must occur even in random noise.
Consider this example (*not* meant as an analogy for the discrete math nazis): you have an infinite sequence of completely random letters over the alphabet. What is the probability of finding "abc" repeated 15 times with a gap of exactly 10 letters between successive repeats? If the stream is indeed completely random, then the probability is non-zero and you will EVENTUALLY (probably) see the "pattern".
This sounds like they're setting themselves up for one of the less predictable problems with the expert systems of the '80s: namely the fact that experts sometimes don't want to transfer their hard-acquired knowledge into a box designed to replace them. But given that this is the Air Force, "orders" might be the solution.
Geek 2:
And they're both absolutely right. I would suggest starting with a free template and modifying the CSS / graphics. That saves you the initial legwork of choosing a design layout, colors, etc. Here are some sites with some really awesome templates and liberal licensing (CC for most I think): Free CSS Templates.org and Open source web design.
This is *exactly* the problem with this branch of computational intelligence, stuff that you see at any CI/AI conference. PSO is a minor variation of stochastic hill-climbing -- it's a friggin heuristic. There is no guarantee that it will perform better than any of the myriad other "biologically-inspired" heuristic out there, including ACO, GAs, ESs, etc. See the No free lunch theorem.
IMO (and as an earlier poster said I am an AI researcher, albeit not in CV), the whole field is trying to wean off a hangover from '89, when genetic algorithms blew people away with the fact that they were "biologically-inspired" and actually could do optimization. Now every minor heuristic variation of stochastic search has to have some cutesy Disney-like "biologically-inspired" back story -- why?? Just move on, damnit, or try to prove something about your heuristics (then again, see the NFL theorem).
That's all fine, but do we really another idiotic web 2.0 name for a startup? Vlingo?? REALLY!?!? Haven't we had enough of vongo, twitter, oyogi, flickr, xuqa, blinkx, sharkle, squidoo, zemq, diigo, frappr, joost, zingee, vyew, bebo?
aaah...reminds me of college.
Time itself does not slow down, time perception is subjective. Anyone who has ever ingested cannabis and looked at a ticking clock will tell you this. "From what people tell me", it is on the order of "experiencing" 3 seconds every second. No need for free-falling airplanes.
The MS Lifecam is a plain-vanilla webcam. What you're probably thinking of is some of the conjectures of Ray Kurzweil.
This is why the recent BitTorrent lawsuit against Comcast is so important...once they realize that they can't look inside encrypted packets, they're just going to block all p2p traffic. But even that is going to be hard, because at the encrypted UDP packet level, what really distinguishes a BT packet from, say, a Skype packet which is also encrypted by default? Screw encryption, what differentiates a DRM-free MP3 flying in from iTunes or Amazon from one coming through a modified BT protocol which uses port 80 and fake http headers?
In short, this is the dumbest idea and any implementation will be necessarily half-assed and is going to affect people.
SOAB! LiveJournal has been a great source for academic researchers in NLP, social networks and data mining, mainly because they don't mind you crawling their site and users have public "friend lists", etc. I doubt if that culture is going to be retained with a Russian owner. I was at a conference earlier this year, and met a guy from SixApart. He was talking about some sort of open API for crawling multiple social networking sites, but I guess that's probably going to be history now.
Not only that, but we now have some sort of government-manufactured rule-based system that assigns risk to 'potential terrorists'. Just wait for the inevitable leak of their methodology (via stolen laptops, incompetence, etc.) and you just gave real terrorists a way to evade suspicion. That's the problem with any "model" for suspicious behavior -- once its known, it's easily exploited.
It seems like you might want one of these. I believe the new zunes have a wireless dock that grabs podcast subscriptions from your pc while it charges. not sure if the new ipod touch does that, but i'd be surprised if it didn't.
short answer: yes. and even before "refactoring" came into vogue, there were other names for it. hence, TFA is not really a FA.
That's a great knee-jerk reaction. Without understanding the motivation behind the article, you assume that code scavenging means stealing other people's code. What they're really talking about is (legitimately) re-visiting code that you or other people have written, and then picking and modularizing bite-sized chunks. In other words, you would design a large program (mark I) and then go back and pick out useful parts, clean/debug them and have working modules (mark II) for the next project.
Also, for people who haven't read TFA, it's 9 short paragraphs long and barely an article. They talk about a "formal approach to code scavenging" without even coming close to explaining what exactly that MEANS.
Just like you can fly by pulling on your shoelaces, nothing you do can change the weight, either real or "perceived", of an object sitting in your hand.
I absolutely agree with you. Even the Computerworld (admittedly not the pinnacle of scientific reporting) article starts by saying "University of Ottawa student Kris Woodbeck is combining the neural processes we use to understand image data with the features of graphics processors." I don't even know where to begin with that statement. So he's come up with a model of neural image processing (a feat in itself)...and is mapping it to a GPU? This is like saying "we've figured out how to isolate stem cells from a source other than human embryos...and we used plastic petri dishes to do it!."
Second: 'The brain is very parallel. There's lots of things going on at once," he said. "Graphics processors are also very parallel'. OK, is this a science finding (i.e. a new image processing ALGORITHM based on the brain), or a systems paper (we came up with a parallel GPU version of an algorithm). I really hope he was misquoted, because otherwise it sounds like vaporware or untested hypotheses.
And then: 'For images, it might be when you took it, with what camera, with what exposure, that's about it. Then you're stuck with a red barn in rolling hills and I might know it was taken in California, but no one else does. How do you surface that metadata so it becomes much more searchable?' OK, now where did this come from again? neural processing? parallel GPU? and now inferring metadata?
Sorry, but getting a provisional patent is hardly a difficult thing (most universities in the US will file one for free if *YOU* think you might be on to something). Furthermore, all this would be more credible if they published some results and at least a brief description (which is allowable by patent law). Until I see some numbers, this is non-news.
That'll only work if you don't tell them beforehand that they're going to be tested. If you did, only pathological liars would end up lying. And if you didn't, there's not a chance your study would be approved by any sane IRB (institutional review board for human subject trials). Or if you asked people if you could test them only after their response, what are the chances that the liars would take the test (and thus be revealed as "liars")? In all cases, your "lying rate" is very likely to be biased.
In science, if something *plausible* COULD be wrong with a study, it deserves to be analyzed before the study is assumed to be rigorous. This also applies to studies on the other side, i.e. the ones which claim marijuana cures death and stops global warming. In your analogy, you can't possibly give me any plausible evidence that my parents killed JFK. I pointed out a rather common methodological flaw (check the literature) with using self-reporting in smoked marijuana studies.
Would you say there's an honest culture of information about cannabis in the United States? I wouldn't. I don't know if you're being sarcastic or not, but I'd be all for a campaign to educate people on the safe use of marijuana. In the Netherlands, most coffee shops stock a vaporizer and a lot of Dutch people I've talked to would prefer to use a vaporizer. Ultimately, people will probably still smoke cannabis because of the social bonding aspect, but they should be educated about the alternatives. After that, it's a choice you make for yourself.
In case you were NOT being sarcastic, here are some websites that advocate safe marijuana use:
safer choice, regulate, marijuana uses (not really an organization, but an emeritus harvard professor who's studying the positive uses of marijuana)
IANA social scientist, but this seems like its quite difficult to quantify. How can you tell when someone is lying if they say that they do not smoke marijuana? Short of testing their blood/hair/urine/etc or watching them continuously, there's no real objective way to tell if someone smokes cannabis or not. It also seems unlikely to me that someone who DOES agrees to undergo a drug test would lie about their drug use. Furthermore, someone who tests positive for cannabis might not actually SMOKE cannabis, as I mentioned. I also don't see how you could build redundancy into this type of question on a survey (perhaps a psychologist could correct me).
So we're left with only what people report. This is where I tend to turn to experience. In the United States, a lot of cannabis smokers I personally know would probably say "no" to the question on any type of questionnaire, even if it were anonymous. The point is that while you can't verify if someone is lying or not, I can think of a lot of plausible reasons why someone WOULD lie with draconian drug laws.
That said, it's of course still conjecture. That's one reason why I tend to distrust smoked marijuana studies -- they're usually targeting marijuana in its smoked form, not THC as a compound.
This is righteous bullshit. Allow me to elaborate: