If somebody can't tell by the colored box around the sponsored links, or hey, the text that reads "Sponsored Links", then what exactly could Google do to make it more obvious that these results are paid for?
How about a nice animated gif? Or maybe an embedded flash object? After all, isn't intrusiveness what we're really hoping for from web advertising?
Perhaps it is best to think of printing a monochrome image. You do not have "continuous" (i.e. a large number of discrete levels) control over the pixel intensity, unlike a computer monitor where you can directly control intensity of a pixel. Each pixel is binary: either pigment is put down or it is not put down. In order to simulate continuous tones you use digital halftoning, as illustrated. You can think of a 256 level halftone pattern as an 8x8 block of pixels that is filled in algorithmically to give the appropriate level. Halftoning is further complicated by the fact that the pixel size is usually much larger than the addressibility. (A pixel from a 1000dpi printer will generally be much larger than.001 inches. i.e. the pixels overlap considerably.)
You can indeed overlap the three colors, but even if your registration is perfect, you will have at most three bits per pixel when printing with three colors (each color is either present or not). 24 bit color can only be simulated through half-toning, which is essentially overlaying three monochrome half-toned images.
You're still misunderstanding printing technologies. Most of the available technologies (ink jet and related, xerography, thermal transfer, offset lithography) actually print binary pixels. Either a pixel is printed or it is not printed. So you only have as many different color pixels as you have inks (or toners). Digital halftoning (look it up, it's cool) is used to group many pixels together into a much larger dot and color them appropriately to simulate a desired color tone. The limited resolution of the human eye filters out the variation at the pixel level to leave you with the simulated tone. The resolution of the dots is often around 300dpi, even though the pixel resolution (which is usually used in advertising) may be an order of magnitude higher. So your 48 bit color depth with 6 inks really only gives you 6 bits per printed pixel.
The CIA venture capital company is called In-Q-Tel. In-Q-Tel invested in Keyhole, who developed the forerunner to Google Earth. Keyhole ownership converted to Google ownership. I had heard from a reputable source that In-Q-Tel also invested in Google during their early days, though it appears that In-Q-Tel no longer holds any ownership in Google. In-Q-Tel invested in other search companies over the years, though they seem to have divested ownership in most of those companies.
There needn't be anything nefarious about In-Q-Tel funding Google. Remember the explosive growth of the web a decade ago. Before google, you knew there was a ton of stuff out there, but there was no way to find it. Web searches were very hit or miss. Google improved search technology tremendously, and a decent search engine is itself a boon to the intelligence community. People in this discussion have joked about keyword searches for terrorists, but seriously, it's an invaluable tool, even if a Google Maps search for Osama bin Laden doesn't put a little pushpin on the appropriate cave. Hate groups and terrorist networks recruit using the internet. Search technologies make it easer to keep track of what's out there.
And heck, I get some benefit from decent search technology too.
Re:Why software sucks in one sentance
on
Why Software Sucks
·
· Score: 1
Does no one remember this from 2002?
A Chinese newspaper, the Beijing Evening News, picked up an Onion article and ran it as an authentic story. The article said that Congress threatened to leave DC unless a new capitol was built. The retraction is posted here, and includes some classic lines, like
Some small American newspapers often compose novelty news stories, to attract people's attention and make money. The New York-based weekly "The Onion" is just such a paper.
According to people working at Capitol, the Onion is a paper which is constantly reporting untrue stories.
So apparently it is possible for people to be misled...
What robot meant originally? You mean "worker" in Czech? From the play?
In the robotics lab where I worked, the generally accepted definition for robot was "a programmable source of work," where work is in the physics sense of forces and movement. This definition will include most things you would think of as robots, but would also include devices such as automatic breadmakers. I find this a useful, if broad, definition.
Industrial robots have a long history and are generally accepted to be robots (at least by people who call themselves roboticists). Industrial robots are often autonomous, but rarely make decisions or have AI of any kind. (An industrial robot is a more general device than special purpose automation. The manufacturing line for pampers will be automated, but does not necessarily contain robots.)
Under the above definition, the passive walkers described in the Science paper would not be considered robots because they are not programmable. But the powered walkers (basically a passive walker with an actuator or two added somewhere) would indeed be considered robots.
According to a theorem usually attributed to Cybenko, any continuous nonlinear function can be represented by a linear combination of sigmoid functions of a linear combination of your parameters. In neural nets terms, a single hidden layer net with 2n+1 neurons in your hidden layer can represent _any_ continuous function.
If you're talking about Cybenko's '89 paper, I believe you misunderstand the result. The main theorem simply says that for any continuous function f on the interval and any epsilon>0, there exists a "neural network" (hidden layer of n sigmoidal neurons) that is epsilon close to f in the infinity norm. So even if you had a training technique that could avoid all local minima, Cybenko's result does not indicate the number of neurons required to attain a certain epsilon. No guarantee of the quality of approximation is given for fixed n. (This is not surprising if you think through the case n=1,2.)
Cybenko's result was more of a reality check for the NN community than anything else. If NNs didn't have this property, there wouldn't be much use in studying them. The Weierstrass Approximation theorem, which you can find in a good real analysis book, shows that plain old vanilla polynomials of the form sum(i=1,n) a_i x^i have the same property.
Barron had a paper giving rate-of-approximation results for a certain class of functions. This starts to answer the question "how big should n be?" I'm not sure what new work has appeared along these lines. I've been out of touch with the NN community for a few years. That said, a lot of the learning people seem to be more excited about support vector machines and kernel methods these days. I guess some people group these techniques along with neural networks, but they lie on a much more solid foundation of statistics.
Superconductivity is a phenomenon occurring in certain materials at low temperatures, characterised by the complete absence of electrical resistance and the damping of the interior magnetic field (the Meissner effect.) ..and... Experiments have in fact demonstrated that currents in superconducting rings persist for years without any measurable degradation.
A number of people have suggested using those ridged things that give different images from different angles. But even if you could give a unique image from each angle, that would not be enough...
For a fixed angle, if an observer moves twice as far away from the invisibilty screen, the screen will now subtend a smaller angle, and thus must project a different image.
This problem seems more difficult to overcome.
Hilbert's problems are very well known among mathematical types. Check out the book
The Honors Class, written for a general audience, which covers the problems and the mathematicians who solved them. (Mathematicians can be a wacky bunch, so the bios make an interesting read. The problems themselves are cool too.)
No really, I'm serious. This robot's playing style is more akin to what we call "whackball." Better foosball players remain in contact with the ball a lot longer, "palming" the ball with the face of the man to do quick changes in direction, fakes, and more. There's some video available here on the right side of the page.
Of course, there have been robots interacting with dynamic environments in similar ways for a long time, such as juggling and running. It's a big jump to go to the next level, which requires chaining together sequences of difficult actions, such as palming the ball, passing, and shooting.
But I think foosball is a great place to explore such dynamical interaction and action composition, and I'm jealous that they beat me to building a table.
Where did you get that information on chopsticks? It sounded like interesting story to have on hand, so I took a look at Enclopedia Brittanica, which said:
"Chopsticks of bamboo or wood, and subsequently of ivory and precious metals, originated in China as early as the Shang dynasty (c. 1766-c. 1122 BC) and from there spread throughout East Asia. In China, the substitution of chopsticks for knives at the table reflected the ascendancy of the scholar over the warrior as a cultural hero."
Which would seem to indicate that chopsticks were around for several millenia, and also gives some basis for the pretension associated with them. Do you have any further references on the use of chopsticks in mining restaurants in the U.S.?
Puns were considered the lowest form of wit in Enlightenment France (Voltaire and all those guys). Watch the movie Ridicule (which is really good) to get the picture.
Pool started out to write a book on nuclear technology for the Sloan book series, but ended up with a very broad case study that includes betamax vs. VHS, internal combustion vs. steam cars, and AC vs. DC power (i.e. why DC power was in use as long as it was.). Nuclear technology is still a running theme in the book.
There are lots of studies using fMRI (functional MRI,the more appropriate name for what they are using, since they actually take a "movie" of MRI snapshots in the brain.) The fields that use fMRI most heavily is neuroscience and cognitive science.
As an example of studies I'm aware of, people are monitored while doing simple tasks, such as reading the words on a series of cards. The words are color names, and sometimes the color name is written in the same color as the name and sometimes in a different color. Simple motor tasks, such as squeezing a ball, are also examined.
The drawbacks of fMRI is that they only allow you to monitor bloodflow in the brain. The idea is that areas that are being actively used will get increased blood flow, but this happens on the order of tens of seconds, so it doesn't really provide a detailed picture of brain activity.
The brain is immensely complicated, and it will be a long time before we have even a simple understanding of how it functions. Even longer till we understand higher level tasks, such as how we decide to buy something.
I, like one of the critics in the article, suspect that this "neuromarketing" effort will prove ineffective, at least in the short run. The biggest worry would be that they'd stuble onto something like the "orienting response" that keeps people in front of the TV.
Yup, the company was GO, and the book about it is:
Startup
by Jerry Kaplan
Kaplan had the idea for pen computing and founded GO to pursue the dream. The book is based on his personal diary and gives a pretty good view of starting a company, seeking capital, expansion, and ultimately failing. (And this was a good idea, not a dot.com) You get to see some nasty moves by a number of the other players, including Microsoft, Apple, and (I believe) Intel, among others. I highly recommend the book!
The Pathfinder mission was widely regarded as a huge success. Several days into the mission the system began restarting intermittently. The bug was located and corrected, and is a great lesson in real-time operating systems, priority-based preemptive scheduling, and (the fix) priority inversion.
There's lots of information on the web, and it makes a good read, even if you're not into operating systems.
For example What really happened on Mars Introduction to Priority Inversion"
"I, Cringely" has an interesting article here on rolling your own DSL connection. The basic idea is to rent a "dry copper pair," basically wires you can use for whatever you want. If someone on your side of your telephone companies Central Office has broadband available, this might be something to seriously look into.
Slashdot Mirror
How about a nice animated gif? Or maybe an embedded flash object? After all, isn't intrusiveness what we're really hoping for from web advertising?
Perhaps it is best to think of printing a monochrome image. You do not have "continuous" (i.e. a large number of discrete levels) control over the pixel intensity, unlike a computer monitor where you can directly control intensity of a pixel. Each pixel is binary: either pigment is put down or it is not put down. In order to simulate continuous tones you use digital halftoning, as illustrated. You can think of a 256 level halftone pattern as an 8x8 block of pixels that is filled in algorithmically to give the appropriate level. Halftoning is further complicated by the fact that the pixel size is usually much larger than the addressibility. (A pixel from a 1000dpi printer will generally be much larger than .001 inches. i.e. the pixels overlap considerably.)
You can indeed overlap the three colors, but even if your registration is perfect, you will have at most three bits per pixel when printing with three colors (each color is either present or not). 24 bit color can only be simulated through half-toning, which is essentially overlaying three monochrome half-toned images.
You're still misunderstanding printing technologies. Most of the available technologies (ink jet and related, xerography, thermal transfer, offset lithography) actually print binary pixels. Either a pixel is printed or it is not printed. So you only have as many different color pixels as you have inks (or toners). Digital halftoning (look it up, it's cool) is used to group many pixels together into a much larger dot and color them appropriately to simulate a desired color tone. The limited resolution of the human eye filters out the variation at the pixel level to leave you with the simulated tone. The resolution of the dots is often around 300dpi, even though the pixel resolution (which is usually used in advertising) may be an order of magnitude higher. So your 48 bit color depth with 6 inks really only gives you 6 bits per printed pixel.
The CIA venture capital company is called In-Q-Tel. In-Q-Tel invested in Keyhole, who developed the forerunner to Google Earth. Keyhole ownership converted to Google ownership. I had heard from a reputable source that In-Q-Tel also invested in Google during their early days, though it appears that In-Q-Tel no longer holds any ownership in Google. In-Q-Tel invested in other search companies over the years, though they seem to have divested ownership in most of those companies.
There needn't be anything nefarious about In-Q-Tel funding Google. Remember the explosive growth of the web a decade ago. Before google, you knew there was a ton of stuff out there, but there was no way to find it. Web searches were very hit or miss. Google improved search technology tremendously, and a decent search engine is itself a boon to the intelligence community. People in this discussion have joked about keyword searches for terrorists, but seriously, it's an invaluable tool, even if a Google Maps search for Osama bin Laden doesn't put a little pushpin on the appropriate cave. Hate groups and terrorist networks recruit using the internet. Search technologies make it easer to keep track of what's out there.
And heck, I get some benefit from decent search technology too.
JPL also had a hiccup with priority inversion on Pathfinder.
A Chinese newspaper, the Beijing Evening News, picked up an Onion article and ran it as an authentic story. The article said that Congress threatened to leave DC unless a new capitol was built. The retraction is posted here, and includes some classic lines, like So apparently it is possible for people to be misled...
What robot meant originally? You mean "worker" in Czech? From the play?
In the robotics lab where I worked, the generally accepted definition for robot was "a programmable source of work," where work is in the physics sense of forces and movement. This definition will include most things you would think of as robots, but would also include devices such as automatic breadmakers. I find this a useful, if broad, definition.
Industrial robots have a long history and are generally accepted to be robots (at least by people who call themselves roboticists). Industrial robots are often autonomous, but rarely make decisions or have AI of any kind. (An industrial robot is a more general device than special purpose automation. The manufacturing line for pampers will be automated, but does not necessarily contain robots.)
Under the above definition, the passive walkers described in the Science paper would not be considered robots because they are not programmable. But the powered walkers (basically a passive walker with an actuator or two added somewhere) would indeed be considered robots.
Cybenko's result was more of a reality check for the NN community than anything else. If NNs didn't have this property, there wouldn't be much use in studying them. The Weierstrass Approximation theorem, which you can find in a good real analysis book, shows that plain old vanilla polynomials of the form sum(i=1,n) a_i x^i have the same property.
Barron had a paper giving rate-of-approximation results for a certain class of functions. This starts to answer the question "how big should n be?" I'm not sure what new work has appeared along these lines. I've been out of touch with the NN community for a few years. That said, a lot of the learning people seem to be more excited about support vector machines and kernel methods these days. I guess some people group these techniques along with neural networks, but they lie on a much more solid foundation of statistics.
cheers, Rick
From Wikipedia,
Superconductivity is a phenomenon occurring in certain materials at low temperatures, characterised by the complete absence of electrical resistance and the damping of the interior magnetic field (the Meissner effect.)
..and...
Experiments have in fact demonstrated that currents in superconducting rings persist for years without any measurable degradation.
That seems pretty close to zero.
A number of people have suggested using those ridged things that give different images from different angles. But even if you could give a unique image from each angle, that would not be enough...
For a fixed angle, if an observer moves twice as far away from the invisibilty screen, the screen will now subtend a smaller angle, and thus must project a different image. This problem seems more difficult to overcome.
Hilbert's problems are very well known among mathematical types. Check out the book The Honors Class, written for a general audience, which covers the problems and the mathematicians who solved them. (Mathematicians can be a wacky bunch, so the bios make an interesting read. The problems themselves are cool too.)
No really, I'm serious. This robot's playing style is more akin to what we call "whackball." Better foosball players remain in contact with the ball a lot longer, "palming" the ball with the face of the man to do quick changes in direction, fakes, and more. There's some video available here on the right side of the page.
Of course, there have been robots interacting with dynamic environments in similar ways for a long time, such as juggling and running. It's a big jump to go to the next level, which requires chaining together sequences of difficult actions, such as palming the ball, passing, and shooting. But I think foosball is a great place to explore such dynamical interaction and action composition, and I'm jealous that they beat me to building a table.
Hi,
Where did you get that information on chopsticks? It sounded like interesting story to have on hand, so I took a look at Enclopedia Brittanica, which said:
"Chopsticks of bamboo or wood, and subsequently of ivory and precious metals, originated in China as early as the Shang dynasty (c. 1766-c. 1122 BC) and from there spread throughout East Asia. In China, the substitution of chopsticks for knives at the table reflected the ascendancy of the scholar over the warrior as a cultural hero."
Which would seem to indicate that chopsticks were around for several millenia, and also gives some basis for the pretension associated with them. Do you have any further references on the use of chopsticks in mining restaurants in the U.S.?
Puns were considered the lowest form of wit in Enlightenment France (Voltaire and all those guys). Watch the movie Ridicule (which is really good) to get the picture.
is Beyond Engineering: How Society Shapes Technology by Robert Pool.
Pool started out to write a book on nuclear technology for the Sloan book series, but ended up with a very broad case study that includes betamax vs. VHS, internal combustion vs. steam cars, and AC vs. DC power (i.e. why DC power was in use as long as it was.). Nuclear technology is still a running theme in the book.
Another great science search engine (though not a digital library) is Scirus.
There are lots of studies using fMRI (functional MRI,the more appropriate name for what they are using, since they actually take a "movie" of MRI snapshots in the brain.) The fields that use fMRI most heavily is neuroscience and cognitive science.
As an example of studies I'm aware of, people are monitored while doing simple tasks, such as reading the words on a series of cards. The words are color names, and sometimes the color name is written in the same color as the name and sometimes in a different color. Simple motor tasks, such as squeezing a ball, are also examined.
The drawbacks of fMRI is that they only allow you to monitor bloodflow in the brain. The idea is that areas that are being actively used will get increased blood flow, but this happens on the order of tens of seconds, so it doesn't really provide a detailed picture of brain activity.
The brain is immensely complicated, and it will be a long time before we have even a simple understanding of how it functions. Even longer till we understand higher level tasks, such as how we decide to buy something.
I, like one of the critics in the article, suspect that this "neuromarketing" effort will prove ineffective, at least in the short run. The biggest worry would be that they'd stuble onto something like the "orienting response" that keeps people in front of the TV.
No, it's only a misdemeanor so long as the advertised products sum to less than $2000!
Yup, the company was GO, and the book about it is:
Startup
by Jerry Kaplan
Kaplan had the idea for pen computing and founded GO to pursue the dream. The book is based on his personal diary and gives a pretty good view of starting a company, seeking capital, expansion, and ultimately failing. (And this was a good idea, not a dot.com) You get to see some nasty moves by a number of the other players, including Microsoft, Apple, and (I believe) Intel, among others.
I highly recommend the book!
The Pathfinder mission was widely regarded as a huge success. Several days into the mission the system began restarting intermittently. The bug was located and corrected, and is a great lesson in real-time operating systems, priority-based preemptive scheduling, and (the fix) priority inversion.
There's lots of information on the web, and it makes a good read, even if you're not into operating systems.
For example
What really happened on Mars
Introduction to Priority Inversion"
For the physics behind lasers, transistors, and other semiconductor marvels, check out the Britney Spears Guide to Semiconductor Physics!
"I, Cringely" has an interesting article here on rolling your own DSL connection. The basic idea is to rent a "dry copper pair," basically wires you can use for whatever you want. If someone on your side of your telephone companies Central Office has broadband available, this might be something to seriously look into.