its reasonable to assume that the pattern of defects well-approximated by a poisson distribution, so they will clump. contrary to intuition, "evenly spaced" is highly non-random (it implies dependence between events). iirc i think there is also some shared circuitry between rows of sensors so a defect might knock out an entire row of pixels or some part thereof. another issue is that the defects, being byproducts of chemical impurities, may not necessarily shrink with the circuitry, i.e., having 10 times the resolution would result in each defect affecting 10 times the number of pixels (maybe someone who knows more about the fab process can confirm this...).
ML is basically that the essentially linear minaturization of circuits, when packed into 2 dimensions, provides an squared increase in capacity. (if it was feasible to make cubes of silicon then ML would have a cubic growth rate). this applies to the improvements in camera sensors, but only up to a certain point -- there are some limiting factors...
first, thermal noise and leakage in digital sensors has a dramatic influence on image quality, and these things do not necessarily decrease with miniturization, so image quality does not necessarily improve because to get a good image in the end you have to apply noise reduction which means reducing the spatial resolution with a blur. (for example, some notorious 8 megapixel sensors take worse pictures than 5 mp sensors with the same size).
second, all lenses, existing in the real world and having aperatures of finite size, function as a low pass filter on the image signal (i.e. they blur things slightly even when in perfect focus). the quality of the lens therefore puts an effective limitation on the maximum spatial resolution of *any" type of sensor (film included) of a particular size. the only way to improve this situation is to increase the size of the imaging sensor (e.g., medium and large format film). making this happen for digital cameras involves some rather thorny manufacturing problems which will probably be solved eventually, but are not related to ML. (this is sort of similar to the quantum limitation of ML as well)
What concerns me is the broader trend of women shying away from the analytical diciplines (math, cs, physics, etc). Not particularly because they should be entering those fields for a career but because these fields are representative of an important skill to develop that is really essential regardless of what field one eventually works in. And, noting that this trend is not universal (e.g., consider Indian women in Math/CS) its clear that this trend is a result of social bias. Even if one does not consider the morality issue (e.g., fairness, equality of power, etc), we should be concerned about this because it indicates that our educational system is substandard and therefore producing a workforce which is less than ideal.
So why does nobody make a peep about male discrimination? Society generally seems to prefer a "minimax" approach to dealing with inequity compensation. i.e., the people with the biggest problem get dealt with first (at least with respect to the "event horizon" of a given society which may be blinded to the suffering of certain subcultures or outsiders). While it undoubtedly indicates some sort of inequity, it is hard to argue that the lack of men in nursing or as english professors represents any kind of serious socio-economic problem.
In my opinion gender inequities in both directions are likely related to a common underlying issue. How this is dealt with might be analogous to trying to solve an equation with respect to some variable; you can start on either side, so you start with the one that intuitively looks like it will get you to the answer fastest. I think that translates to a prioritization of the "women's perspective" since it is really a bit more obvious how to proceed and the results have a clear and immediate payoff.
in most of america raising taxes is the politicians' certain death. thus the need to go to great lengths to obtain the necessary resources through increasingly convoluted inventions such as this one.
Re:No one notices a well done security job...
on
Security's Shaky State
·
· Score: 2, Interesting
Suppose we force companies to pay reasonable damages (no criminal charges or anything unless criminal negligence is provable). Naturally, they can and will get liability insurance to cover this, and the actuaries will determine how much that will cost on the basis of how risky their operation is. Similar to having airbags in your car, companies will qualify for discounts by using known secure systems and hiring competent IT security staff. Software/hardware vendors will be motivated to produce secure products because otherwise they will lose business.
Now, in the end the cost gets pushed out to the consumer anyways; so we end up paying for it one way or another -- either you get identity theft insurance to help you deal with the inevitable breach or you force the companies to get insurance or otherwise take appropriate measures. I think the latter option is more efficient because it attacks the problem at the source. Furthermore, we can make the insurance mandatory -- just like driving a motor vehicle, handling peoples' private information puts the general public at risk.
Whatever happens we can't just let this nonsense continue unchecked. I guess HIPPA, Sarbanes and some other laws are going to start dealing with this, but I have yet to see if those have any respectable teeth. And from what I've heard first hand about some of the new systems going into some not-to-be-named HMOs, I don't have a lot of hope that things will get better.
Maybe its a problem of insufficient penalty. The legal consequences regarding breaches of private information seem pretty lax (are there any at all?). Morality is meaningless in the boardroom. The desired behavior has to be brought into compliance by either the market or the government in some form.
...which is why the fire dept responds to medical emergencies also, they are trained to do and can get there sooner due to the higher density of fire stations. They probably won't transport you to the hospital but they can stablize the situation.
In California this is a law and the number of cars is 5. Of course, there has to be a reasonable place to pull over. On roads where this sort of thing is likely to be an issue there are pullouts every couple of miles constructed and signs instructing slower traffic to use pullouts.
People are ignorant but not reckless, which is why the distribution of speed does not change significantly when the speed limit is placed above the 85th percentile. On the other hand placing the limit too low results in more non-compliance, and therefore greater variance in the distribution (i.e. the spread between the non-compliant drivers and the compliant drivers). Greater variance means less predictability in the flow of traffic which results in more accidents.
Another way to look at this is that the ignorance of drivers is basically independent of the posted speed limit. All you can do is make the situation worse by posting the wrong limit. There are lots of ways to make people better drivers, but just lowering the speed limit is not one of them. Speed cameras, GPS systems, etc might actually be a viable solution. Unfortunately the whole speed regulation system has been rife with corruption for years, so there is understandably a great deal of public distrust.
Diebold picked up the e-voting stuff fairly recently when they bought some company (called Electronic Voting Systems or something). In fact some of the murky coverups had already happened before they were owned by Diebold. Diebold may have plenty of good engineers, they just had a major lapse in management when they bought out this two-bit company, their crap-product and sloppy engineers.
AFAIK 20khz is about the upper limit for human hearing at any age. Animals such as cats can hear up to about 30khz. According to some experimental data, lab animals exposed to noise causing a temporary hearing loss (e.g., fully recovered after one week) suffered severe loss much later in life. This would tend to imply that teens exposed to excessive noise will not suffer the consequence until many years later.
Some simple statistical analysis can tell you how much bandwidth is actually needed to service a large number of users. Assuming that network traffic is Poisson distributed (which is more or less true within a given hour of the day), the probability of packet collision falls off exponentially. Because the big telecom companies hire statisticians to figure this stuff out, you can be certain that the total bandwidth out of the neighborhood is significantly less than 500 * bandwidth to each house, and yet the quality of the connection does not suffer (at least, not so much that people will complain). Incidentally the same principle is used with deployment of phone switching equipment, i.e. the number of actual lines going into an area is less than the actual number of lines sold, which is why the phone system jams up rather quickly when a non-Poisson event happens (e.g. an earthquake or a terrorist attack).
i disagree to some extent; its true that top notch education can be had overseas. but many of those countries with "higher standards" are relaxing their standards now, because, suprisingly, the severity of their program scares away their best people. top notch science comes out of research, and risk-taking. rigorous academic standards broadcast from the ivory tower are antithical to innovation. too much of a good thing can backfire.
Yes, the environment can adapt and recover, but really the problems that global warming entails are problems for *humans* -- primarily issues of health and economics.
With an average hit rate of 3 per second, approximatly once in 100 seconds the hit rate will meet or exceed 8 hits per second. So to ensure a smooth experience 99% of the time the server needs to have about three times the 3-hits-per-second capability on hand. This does not take into account the slashdot effect, which is not Poisson distributed.
Slashdot Mirror
its reasonable to assume that the pattern of defects well-approximated by a poisson distribution, so they will clump. contrary to intuition, "evenly spaced" is highly non-random (it implies dependence between events). iirc i think there is also some shared circuitry between rows of sensors so a defect might knock out an entire row of pixels or some part thereof. another issue is that the defects, being byproducts of chemical impurities, may not necessarily shrink with the circuitry, i.e., having 10 times the resolution would result in each defect affecting 10 times the number of pixels (maybe someone who knows more about the fab process can confirm this...).
ML is basically that the essentially linear minaturization of circuits, when packed into 2 dimensions, provides an squared increase in capacity. (if it was feasible to make cubes of silicon then ML would have a cubic growth rate). this applies to the improvements in camera sensors, but only up to a certain point -- there are some limiting factors...
first, thermal noise and leakage in digital sensors has a dramatic influence on image quality, and these things do not necessarily decrease with miniturization, so image quality does not necessarily improve because to get a good image in the end you have to apply noise reduction which means reducing the spatial resolution with a blur. (for example, some notorious 8 megapixel sensors take worse pictures than 5 mp sensors with the same size).
second, all lenses, existing in the real world and having aperatures of finite size, function as a low pass filter on the image signal (i.e. they blur things slightly even when in perfect focus). the quality of the lens therefore puts an effective limitation on the maximum spatial resolution of *any" type of sensor (film included) of a particular size. the only way to improve this situation is to increase the size of the imaging sensor (e.g., medium and large format film). making this happen for digital cameras involves some rather thorny manufacturing problems which will probably be solved eventually, but are not related to ML. (this is sort of similar to the quantum limitation of ML as well)
Uhm, so why exactly does the sci/vis prefer the g5 platform?
"I brake for ultra-thin laptops"
the penis-worms?
My god man, you've just written an infinite loop!
What concerns me is the broader trend of women shying away from the analytical diciplines (math, cs, physics, etc). Not particularly because they should be entering those fields for a career but because these fields are representative of an important skill to develop that is really essential regardless of what field one eventually works in. And, noting that this trend is not universal (e.g., consider Indian women in Math/CS) its clear that this trend is a result of social bias. Even if one does not consider the morality issue (e.g., fairness, equality of power, etc), we should be concerned about this because it indicates that our educational system is substandard and therefore producing a workforce which is less than ideal.
So why does nobody make a peep about male discrimination? Society generally seems to prefer a "minimax" approach to dealing with inequity compensation. i.e., the people with the biggest problem get dealt with first (at least with respect to the "event horizon" of a given society which may be blinded to the suffering of certain subcultures or outsiders). While it undoubtedly indicates some sort of inequity, it is hard to argue that the lack of men in nursing or as english professors represents any kind of serious socio-economic problem.
In my opinion gender inequities in both directions are likely related to a common underlying issue. How this is dealt with might be analogous to trying to solve an equation with respect to some variable; you can start on either side, so you start with the one that intuitively looks like it will get you to the answer fastest. I think that translates to a prioritization of the "women's perspective" since it is really a bit more obvious how to proceed and the results have a clear and immediate payoff.
in most of america raising taxes is the politicians' certain death. thus the need to go to great lengths to obtain the necessary resources through increasingly convoluted inventions such as this one.
Suppose we force companies to pay reasonable damages (no criminal charges or anything unless criminal negligence is provable). Naturally, they can and will get liability insurance to cover this, and the actuaries will determine how much that will cost on the basis of how risky their operation is. Similar to having airbags in your car, companies will qualify for discounts by using known secure systems and hiring competent IT security staff. Software/hardware vendors will be motivated to produce secure products because otherwise they will lose business.
Now, in the end the cost gets pushed out to the consumer anyways; so we end up paying for it one way or another -- either you get identity theft insurance to help you deal with the inevitable breach or you force the companies to get insurance or otherwise take appropriate measures. I think the latter option is more efficient because it attacks the problem at the source. Furthermore, we can make the insurance mandatory -- just like driving a motor vehicle, handling peoples' private information puts the general public at risk.
Whatever happens we can't just let this nonsense continue unchecked. I guess HIPPA, Sarbanes and some other laws are going to start dealing with this, but I have yet to see if those have any respectable teeth. And from what I've heard first hand about some of the new systems going into some not-to-be-named HMOs, I don't have a lot of hope that things will get better.
Maybe its a problem of insufficient penalty. The legal consequences regarding breaches of private information seem pretty lax (are there any at all?). Morality is meaningless in the boardroom. The desired behavior has to be brought into compliance by either the market or the government in some form.
I swear to God, I'll fly a jet into the BellSouth tower...
;)
Ahem. Might want to post AC next time...
Because progress does not buy mansions and private jets. The rotten apples are on the top of the barrel.
...which is why the fire dept responds to medical emergencies also, they are trained to do and can get there sooner due to the higher density of fire stations. They probably won't transport you to the hospital but they can stablize the situation.
In California this is a law and the number of cars is 5. Of course, there has to be a reasonable place to pull over. On roads where this sort of thing is likely to be an issue there are pullouts every couple of miles constructed and signs instructing slower traffic to use pullouts.
Just curious; what is the point of using the handbrake (in lieu of normal brakes) while stopped at a light?
People are ignorant but not reckless, which is why the distribution of speed does not change significantly when the speed limit is placed above the 85th percentile. On the other hand placing the limit too low results in more non-compliance, and therefore greater variance in the distribution (i.e. the spread between the non-compliant drivers and the compliant drivers). Greater variance means less predictability in the flow of traffic which results in more accidents.
Another way to look at this is that the ignorance of drivers is basically independent of the posted speed limit. All you can do is make the situation worse by posting the wrong limit. There are lots of ways to make people better drivers, but just lowering the speed limit is not one of them. Speed cameras, GPS systems, etc might actually be a viable solution. Unfortunately the whole speed regulation system has been rife with corruption for years, so there is understandably a great deal of public distrust.
Diebold picked up the e-voting stuff fairly recently when they bought some company (called Electronic Voting Systems or something). In fact some of the murky coverups had already happened before they were owned by Diebold. Diebold may have plenty of good engineers, they just had a major lapse in management when they bought out this two-bit company, their crap-product and sloppy engineers.
AFAIK 20khz is about the upper limit for human hearing at any age. Animals such as cats can hear up to about 30khz. According to some experimental data, lab animals exposed to noise causing a temporary hearing loss (e.g., fully recovered after one week) suffered severe loss much later in life. This would tend to imply that teens exposed to excessive noise will not suffer the consequence until many years later.
Some simple statistical analysis can tell you how much bandwidth is actually needed to service a large number of users. Assuming that network traffic is Poisson distributed (which is more or less true within a given hour of the day), the probability of packet collision falls off exponentially. Because the big telecom companies hire statisticians to figure this stuff out, you can be certain that the total bandwidth out of the neighborhood is significantly less than 500 * bandwidth to each house, and yet the quality of the connection does not suffer (at least, not so much that people will complain). Incidentally the same principle is used with deployment of phone switching equipment, i.e. the number of actual lines going into an area is less than the actual number of lines sold, which is why the phone system jams up rather quickly when a non-Poisson event happens (e.g. an earthquake or a terrorist attack).
by replacing them with cheap imports from mexico? brilliant idea. who comes up with this stuff?
i disagree to some extent; its true that top notch education can be had overseas. but many of those countries with "higher standards" are relaxing their standards now, because, suprisingly, the severity of their program scares away their best people. top notch science comes out of research, and risk-taking. rigorous academic standards broadcast from the ivory tower are antithical to innovation. too much of a good thing can backfire.
I'd rather see MySQL AB get their money than Boies and associates.
Yes, the environment can adapt and recover, but really the problems that global warming entails are problems for *humans* -- primarily issues of health and economics.
With an average hit rate of 3 per second, approximatly once in 100 seconds the hit rate will meet or exceed 8 hits per second. So to ensure a smooth experience 99% of the time the server needs to have about three times the 3-hits-per-second capability on hand. This does not take into account the slashdot effect, which is not Poisson distributed.
Lecture 1a: History of OSS ... ... ...
Lecture 3b: Why SCO sucks.
Lecture 3c: Implications of forking.
Lecture 9a; How to use IRC
Could be interesting for people who don't already know these things.