Which, indeed, they did not: 'Although you can never prove that correlation is equal to causation, certainly the most plausible explanation is that [the tremors] are related to the gas injection.'
In fact, they took the very valid point that coincidence (not even correlation, as CrimsonAvenger correctly notes that other seemingly similar cases do not display the same coincidence) does not imply causation, and then decided to breeze past it and declare that "certainly" that causation is the "most plausible explanation". In other words, coincidence --> correlation --> causation. I don't dispute that observation could be used to prove this causation, but where are those observations?
IMHO, just throwing more sensors at a city does not make it smarter. What city infrastructures are really missing now is smart data analytics to make sense of raw sensor data. Such analytical technology is only beginning to emerge (for example, in the space I know best, my own company TaKaDu which provides software to analyse the data from a rather sparse array of sensors in local water distribution networks, and monitor for faults). Without that, you end up with ranks of analysts and engineers staring at columns of numbers which go up and down for a long list of reasons, hunting for the few meaningful patterns or anomalies, with predictably limited results. "More data" or even "more mixing of different data" is just the first stepping stone.
On a related note, an interesting industry collaboration on these topics (again, within the water industry) is SWAN -- Smart Water Networks Forum, focused on the wide variety of data systems which go into managing water networks, one of the more hidden, but most critical, city infrastructures.
If I already have my smartphone with me, or indeed any phone with camera, and I can take pictures of my skin condition (or whatever), and write or talk about it, perhaps I could SEND the photos and comments to a REAL LIVE doctor, even one who works more than a few hours drive away, but within phone coverage, and get his NATURAL INTELLIGENCE comments, instructions, etc. over the phone? Is AI going to be better and cheaper than human doctors any time soon? And assuming it does get that way, why does it need to run on my smartphone? Through the magic that is the interweb (of which I believe smartphones are a part), would it not be easier to send the data to a slightly more powerful SERVER, and get my AI GP's advice from there? Slapping "AI" and "smartphone" on a problem does not make for a brilliant futuristic solution.
I tried reading your comment carefully, honest. I even focused on words 7-10, "hashes relying on is", because they seemed to almost make sense. But I'm afraid the bottom line is that there just isn't enough English, sense, and/or cryptography in that paragraph.
Let's go back to the 1950s, then, and change the accepted nomenclature CS people have been using ever since then, shall we? Because this sounds ever so slightly like something else someone may have wanted to call a "genetic algorithm".
Tell you what, let's go the whole hog, and decide that every iterative algorithm with a complex state is called "genetic". And let's also call them "simulated annealing", while we're at it.
Not really. I refer the honourable AC to the already-cited wikipedic reference. "Steepest ascent" is one form of hill-climbing, but just picking some higher-value neighbour is another.
Sorry, but this is hill climbing, pure and simple. The (very cool) result was achieved by introducing random changes ("mutations", if you like) into a "state" or "temporary solution" (the set of polygons), and keeping the new state only if it increases a target function (the similarity to a target image).
The name "genetic algorithm" is actually used for a more complex situation, more reminiscent of our own genetics: the algorithm maintains a pool of states or temporary solutions, selects two (or more) of them with probability proportional to their target-function score, and then randomly recombines them, possibly with "mutations", to generate a new state for the pool. A low-scoring state is probably removed, to keep the pool at constant size.
Quite possibly, a genetic algorithm would do an even better job here, as it could quickly find, for example, two states which each approximates a different half of the image.
Sounds a little like Collaborative Filtering, where other users' ranking of items helps decide what items are suggested to you. Except for the "collaborative" part, of course. Unless I'm stuck in a Memento-like cycle of short-term amnesia, why on earth would I need Google to remind me that I'd rather they showed me item X before item Y? I've already run that particular search, and seen those items... If my rankings got to help someone else at some point, that might be worth the effort.
So, really, this new "Autistic Filtering" framework takes a bold step away from the Social Net fad, instead preferring the Anti-Social Net paradigm. As John Dickinson and countless others did not say, "Divided we stand, united we fall".
I particularly liked the way the video demonstrates g-speak's usefulness for those typical home and "office productivity" tasks I so often find myself getting slowed down by:
- Moving letters around on the screen (especially once they start wiggling and shifting away from you, and let's face it: everyone hates it when that starts) - Rotating squiggles and letter-circles around 3 axes - Navigating through a 3D array of thousands of identical Chinese characters, housed in little boxes (to be fair, though, this sounds a little like the first task)
There was certainly a lot of blogospheric noise about some British agency finding Prozac in drinking water. The only problem: it never happened. Someone hypothesised (wrongly) that this could happen, someone misquoted, and the nonsense took off from there. The actual agency which "found" the Prozac has never actually run such a test. Of course, only the sensational claim is remembered...
Surely high lethality makes for SLOWER contagion?
on
A Flu Pandemic?
·
· Score: 1
Every so often one of these very lethal soon-to-be almost-upon-us epidemics is explained (at least in the popular media) to be a great worldwide risk, because it is highly lethal, and those who contact the disease frequently die within a short period.
This does not make sense. To create a proper epidemic, what you really need is sick people walking around for weeks, coughing, sneezing, or otherwise transmitting their contagious disease. If they die shortly after becoming ill, the personal tragedy remains the same, but they have less chance to pass it on...
Actually, the wonders of modern technology suggest a rather simpler solution. Digital watermarking of video streams is a fairly well-developed field, with several companies offering working products. The "invisible" watermark is some extra bits of "payload" added by some transformation of the images -- nothing which perceptibly degrades image quality -- and can be recovered again by some simple transformation of the data.
Algorithms exist which embed this information "visually", in the sense that it is distirbuted across the whole or much of the image, and it survives "classic" image processing such as resizing, lossy compression, and recolouration of the image (not to any degree, of course, but you'd be ruining the movie before you got rid of the watermark), rather than just being a few specific bits which can be deleted or edited. Some of these techniques are also intended to be tamper-proof, in the sense that without the watermark-creator's key it is very hard to know how to remove or alter the watermark.
Such a watermark would seem to be much better than a glaring visual signal, for tracking down the originator of a leaked copy. It wouldn't stop viewers enjoying their leaked copies, but the leaker could be held accountable.
Slashdot Mirror
Which, indeed, they did not: 'Although you can never prove that correlation is equal to causation, certainly the most plausible explanation is that [the tremors] are related to the gas injection.'
In fact, they took the very valid point that coincidence (not even correlation, as CrimsonAvenger correctly notes that other seemingly similar cases do not display the same coincidence) does not imply causation, and then decided to breeze past it and declare that "certainly" that causation is the "most plausible explanation". In other words, coincidence --> correlation --> causation. I don't dispute that observation could be used to prove this causation, but where are those observations?
IMHO, just throwing more sensors at a city does not make it smarter. What city infrastructures are really missing now is smart data analytics to make sense of raw sensor data. Such analytical technology is only beginning to emerge (for example, in the space I know best, my own company TaKaDu which provides software to analyse the data from a rather sparse array of sensors in local water distribution networks, and monitor for faults). Without that, you end up with ranks of analysts and engineers staring at columns of numbers which go up and down for a long list of reasons, hunting for the few meaningful patterns or anomalies, with predictably limited results. "More data" or even "more mixing of different data" is just the first stepping stone.
On a related note, an interesting industry collaboration on these topics (again, within the water industry) is SWAN -- Smart Water Networks Forum, focused on the wide variety of data systems which go into managing water networks, one of the more hidden, but most critical, city infrastructures.
If I already have my smartphone with me, or indeed any phone with camera, and I can take pictures of my skin condition (or whatever), and write or talk about it, perhaps I could SEND the photos and comments to a REAL LIVE doctor, even one who works more than a few hours drive away, but within phone coverage, and get his NATURAL INTELLIGENCE comments, instructions, etc. over the phone? Is AI going to be better and cheaper than human doctors any time soon? And assuming it does get that way, why does it need to run on my smartphone? Through the magic that is the interweb (of which I believe smartphones are a part), would it not be easier to send the data to a slightly more powerful SERVER, and get my AI GP's advice from there? Slapping "AI" and "smartphone" on a problem does not make for a brilliant futuristic solution.
Yes. And no.
I tried reading your comment carefully, honest. I even focused on words 7-10, "hashes relying on is", because they seemed to almost make sense. But I'm afraid the bottom line is that there just isn't enough English, sense, and/or cryptography in that paragraph.
Let's go back to the 1950s, then, and change the accepted nomenclature CS people have been using ever since then, shall we? Because this sounds ever so slightly like something else someone may have wanted to call a "genetic algorithm". Tell you what, let's go the whole hog, and decide that every iterative algorithm with a complex state is called "genetic". And let's also call them "simulated annealing", while we're at it.
Not really. I refer the honourable AC to the already-cited wikipedic reference. "Steepest ascent" is one form of hill-climbing, but just picking some higher-value neighbour is another.
Sorry, but this is hill climbing, pure and simple. The (very cool) result was achieved by introducing random changes ("mutations", if you like) into a "state" or "temporary solution" (the set of polygons), and keeping the new state only if it increases a target function (the similarity to a target image).
The name "genetic algorithm" is actually used for a more complex situation, more reminiscent of our own genetics: the algorithm maintains a pool of states or temporary solutions, selects two (or more) of them with probability proportional to their target-function score, and then randomly recombines them, possibly with "mutations", to generate a new state for the pool. A low-scoring state is probably removed, to keep the pool at constant size.
Quite possibly, a genetic algorithm would do an even better job here, as it could quickly find, for example, two states which each approximates a different half of the image.
Sounds a little like Collaborative Filtering, where other users' ranking of items helps decide what items are suggested to you. Except for the "collaborative" part, of course. Unless I'm stuck in a Memento-like cycle of short-term amnesia, why on earth would I need Google to remind me that I'd rather they showed me item X before item Y? I've already run that particular search, and seen those items... If my rankings got to help someone else at some point, that might be worth the effort.
So, really, this new "Autistic Filtering" framework takes a bold step away from the Social Net fad, instead preferring the Anti-Social Net paradigm. As John Dickinson and countless others did not say, "Divided we stand, united we fall".
I particularly liked the way the video demonstrates g-speak's usefulness for those typical home and "office productivity" tasks I so often find myself getting slowed down by:
- Moving letters around on the screen (especially once they start wiggling and shifting away from you, and let's face it: everyone hates it when that starts)
- Rotating squiggles and letter-circles around 3 axes
- Navigating through a 3D array of thousands of identical Chinese characters, housed in little boxes (to be fair, though, this sounds a little like the first task)
There was certainly a lot of blogospheric noise about some British agency finding Prozac in drinking water. The only problem: it never happened. Someone hypothesised (wrongly) that this could happen, someone misquoted, and the nonsense took off from there. The actual agency which "found" the Prozac has never actually run such a test. Of course, only the sensational claim is remembered...
http://www.guardian.co.uk/science/2004/aug/12/thisweekssciencequestions3
"That's right, the mod categories are just like the points on 'Whose Line' -- they don't mean anything..."
Hexasigmoidal (adj.)? Hexasigmism (n.)?
Every so often one of these very lethal soon-to-be almost-upon-us epidemics is explained (at least in the popular media) to be a great worldwide risk, because it is highly lethal, and those who contact the disease frequently die within a short period.
This does not make sense. To create a proper epidemic, what you really need is sick people walking around for weeks, coughing, sneezing, or otherwise transmitting their contagious disease. If they die shortly after becoming ill, the personal tragedy remains the same, but they have less chance to pass it on...
Actually, the wonders of modern technology suggest a rather simpler solution. Digital watermarking of video streams is a fairly well-developed field, with several companies offering working products. The "invisible" watermark is some extra bits of "payload" added by some transformation of the images -- nothing which perceptibly degrades image quality -- and can be recovered again by some simple transformation of the data.
Algorithms exist which embed this information "visually", in the sense that it is distirbuted across the whole or much of the image, and it survives "classic" image processing such as resizing, lossy compression, and recolouration of the image (not to any degree, of course, but you'd be ruining the movie before you got rid of the watermark), rather than just being a few specific bits which can be deleted or edited. Some of these techniques are also intended to be tamper-proof, in the sense that without the watermark-creator's key it is very hard to know how to remove or alter the watermark.
Such a watermark would seem to be much better than a glaring visual signal, for tracking down the originator of a leaked copy. It wouldn't stop viewers enjoying their leaked copies, but the leaker could be held accountable.