CS has actually been used in the seismic community since the 70s, seismologists have simply been assuming they could do this because thats all the data they had to work with.
In 2006, Terry Tao and Emmanuel Candes proved that this is true, and described the framework under which it is true.
However, it can't recreate information that wasn't recorded in the first place, so if the audio was recorded at 20kHz you would only get output of audio below 10kHz (the Nyquist frequency [wikipedia.org] in this case), although it's conceivable that even more advanced algorithms could infer these frequencies as most instruments have a predictable distribution of harmonics.
It also seems that most compression algorithms (JPG for example) would destroy these bits of detail that the algorithm would use, so raw data is likely to be needed in most cases. I'm just going off of my knowledge of DSP, I don't know any particulars of this technique beyond this article, but it looks legitimate and very useful as long as you aren't expecting CSI-level miracles.
From what I understand, neither of these statements are in fact the case. Firstly, Compressed sensing is capable of reconstructing data at significantly sub-Nyquist sampling rates, the Nyquist criterion does not apply at all in this case. Secondly, as far as I'm aware, JPG data will work fine, although I think it is possible that some noise may be interpreted as structure, but that is not to say it won't work. In fact that basis in which you construct the data is some sort of fourier, cosine or wavelet basis, which is pretty much what JPG does.
CS works by the assumption that your data is sparse on some basis, that completely rules out the idea that you can simply sample pixel data at random and reconstruct the image. What you have to do is sample data from a sparse basis, essentially you need to take random bits of information from something like the fourier transform or wavelet transform of the image, and what you can reconstruct is the complete set of information about this transformed data set. There is no magic tricky algorithm, just a very very sophisticated proof, in fact the algorithm for compressed sensing (there are many) is any linear programming algorithm (simplex, interior point, etc). Compressed sensing says that if you randomly sample data points from a sparse data set, then the true data set is the most sparse reconstruction you can make that agrees with your sample. This is done by minimising the l1 norm of your data set, given that your reconstruction remains consistent with your sample set. Or more precisely, given an unknown data set x, random sampling matrix A and the resultant sample, b, x can be recovered by the following:
min ||x||_1 such that Ax = b
The logic behind it is completely abstract from any notion that you need to know something about the image to reconstruct it (other than the assumption that it is sparse), you are not so much inferring details of the image from the sampled data as you are saying this: assuming that your data set is sparse, a random selection of that data set contains all the information about its sparsity.
They also, I believe, have the highest educated population in the world. You can't get to where Germany has gotten with a private-sector controlled education system either, so you can score two for Germany.
That said I have heard they have issues with people actually entering the workforce, since so many people stay on at university to complete a PhD, they all end up starting out in the workforce at around 30. Mind you it doesn't seem to be doing too much damage, although they have a smaller per capita gpd than Australia...
The only problem I see with it being adopted by the casual crowd is the fact that it needs to be tethered to a PC. Everyone keeps saying "oh I could easily give this to my mother", but then she still needs a PC with iTunes to run the thing right?
Thats not exactly how it works, the power "assists" your steering, there is always a mechanical linkage so that you can control the car, you just get a boost from the hydrualics. The thing is power steering allows for more favourable suspension geometries which result in heavy steering.
Why is it so dire? Can you give me any credible source of research that says we face utter extinction if we do not stop the CO2 emissions? And what is it with the resources FUD? Where are they going? are we shooting all our resources out into space or something? Matter more or less never changes in any way other than configuration, our mass is going more or less nowhere.
There is a standard for_each, but I think its cumbersome in its own way, and there is a foreach in boost. Heres the boost example code for iterating over a 2d vector:
std::vector<std::vector<int> > matrix_int; BOOST_FOREACH( std::vector<int> & row, matrix_int ) {
BOOST_FOREACH( int & i, row ) {
++i;
} }
Concise, but still somewhat obtuse. The python equiv is still much clearer in its intent, imo. You can just iterate over multidimensional lists in Python and they will be flattened. You can even use Numpy and use slicing syntax to perform the same ends:
import numpy as np
foo = np.array(...) foo += 1 # add 1 to every entry foo[1:] = 2 * foo[:-1] # Update every entry with 2* the previous entry
Its not even that the iterator line is ugly, its the fact that you have to create one in the first place, and its a pointer. I like coding in C++, the output is solid and you feel like you've really achieved something proper, but it is inherently more verbose, obscure, and unreadable than Python.
Python sacrifices a lot for readability, and it really is the most readable language out there.
That is true, but many languages carry with them a lot of housekeeping to get things done, Python tends to keep that sort of stuff out of the domain of the code. Consider a simple example, in C++ say we create a vector and iterate over its elements, we need to make the vector, an iterator, then invoke some pointer arithmetic to actually loop, and then dereference the iterator to extract data. But the important thing here is that we're iterating over items in a list (vector), however you end up with something like:
vector<double> foo = bar; vector<double>::iterator iter = foo.begin(); for(iter; iter != foo.end(); ++iter) {
double thing = *iter;
etc... }
In Python this is simply:
foo = [...] for (item in foo):
thing = item
You can set the C++ out as neatly as you want but the Python is inherently more readable because the housekeeping aspects are removed from the code. I don't care how much of a guru you are, to any human the Python example is infinitely more readable, and more of the code is devoted to describing exactly what is happening, rather than tangential intermediate steps. Yes this means you are trusting Python to do your housekeeping for you, and yes this means a performance hit, but this is besides the point, you shouldn't use Python in an setting where these things are important.
This is not to rag on C++, I really like C++, but it silly to say that Python, probably the most readable language ever invented, is no more readable than any other language. And don't get me started on Java, but then I don't like Java at all.
Exactly how much productive work have you done in Python? Because if you think anyone can output code faster in C than Python, you kind of sound like you've never used python.
"if you think python is the answer, you are kidding yourself."
You'd better tell Google that, maybe they'll start porting all their Python over to PHP?
I think you have the Prius figure a little off, all literature I've read, and accounts from people as well, put it fairly solidy in the 7L/100k range (in realistic terms). To put that into perspective, my 1992 carburettor powered honda civic gets about 7.5L/100k at its peak efficiency.
What MPG (or L/100km as we have down here) does say though is something about *efficiency* at the same time as energy consumption, because we can quite easily relate the notion of gallons (or litres) to our wallet. We can then determine very easily the economic efficiency of our vehicle. This is what is more important to people than outright energy consumption, and this is the only way you can change their mind about consumption habits as well.
Oh and an electric vehicle still consumes resources, so making remarks about infinite MPG is kind of stupid given they are essentially running on stored energy from coal fired steam turbines. So how many miles per unit of coal do you get then?
The government is not running on the assumption that a filter will save all the children, although it certainly is in their PR arsenal. What the government is doing, in their own eyes, is simply closing a loophole in the law. Australia has tough censorship laws *already*, the internet is just not filtered at this point. The government is simply seeking to apply its existing legal framework to the internet.
But this raises the far more important issue: Australia has a draconian censorship framework which needs to be brought into the modern age. The mere fact that the government applying their ratings rules is immediately seen as great wall of China style censorship is indicative of how out of touch the local censorship laws are with contemporary society. This sums up the far bigger problem that critics face: Conroy see this whole issue as applying the law in its intent, the way it is meant to be applied (closing a loophole), in order to get the government to view it any other way would require the government to be convinced it censorship framework needs to be loosened up, R18+ for games would be a nice start!!
Is that any reason to go looking to the state to enforce censorship on an entire country? So that you can have the freedom to not install an internet filter on your own computer?
You speak of a mythical "intelligence" when for all we know we ourselves might well be a collection of simple algorithms being executed in our heads.
No, as I understand it we know fully well that we are not a simple conglomeration of algorithms, isn't that the point of the halting problem? Or incompleteness for that matter (although that is somewhat more esoteric).
Yes a computer is nothing more than a series of algorithms, and this is precisely why it will never be an AI (please note I see an AI as not the mere illusion of intelligence, but true intelligence, created artificially). At least until we can get beyond Turing machines, but then you have to ask if we can even begin to grasp what it means to be intelligent in the first place. And I suspect we don't.
Is any of that actually intelligence though? I see it as sophisticated algorithms, modern advanced mathematics put into practice, and so forth. I don't see much intelligence out of any of the examples you gave and I honestly don't think those sorts of things are what the article was talking about. Autonomy is simply reacting to your environment with a predefined set of instructions. Intelligence is something far more than that.
I'm going to weigh in here and back the GP up. You don't need to stretch the imagination so far as to say this means we have a soul, Kurt Goedel essentially *proved* what this man is trying to say in a cold calculated and highly mathematical manner, back in the 20s. Its depends how you want to interpret the implications of his proof, the actual statement is something to the tune of this: there exists truths about natural numbers which are not provable under any axiomatic system. Fine, not really much to do with the human brain, but this idea is essentially what led Turing to go on a concretely prove that the Halting problem is undecidable. Again, seemingly innocuous? But the implications of these two gigantic theorems are straight forward: Firstly Turing showed without any possibility for argument that a Turing machine can *never* achieve intelligence, more importantly Goedels works shows that the human brain is capable of understanding truths for which we cannot explain systematically, he actually believed this was the core of human intuition.
So how do we proceed from there? We know that we cannot systematically describe everything which the human brain is capable of understanding, so how do you imitate its function if you can't even describe it in the first place?
I am pretty much sure that the current computational models. I.e. Turing Machine are not enough to explain the human mind.
I'm pretty sure you are right, and I'm pretty sure I strongly agree with you. Goedel showed that the human mind is capable of conceiving things beyond the capability of any formal system devised by the human mind, Turing showed that not Turing-complete machine can can solve all problems. Between these two giants and the vast body of work that has grown from their findings, I remain utterly skeptical that AI is even something humans can do. There is something eerily abstract about the whole notion.
I always love to pose this question to people on this topic and similar ones: How can you understand how your mind works from an outside perspective. You can't, its a logical fallacy, everything, absolutely every thought you have ever is the consequence of a human mind at work. You cannot escape that framework and objectively analyse it, and I suspect if you could, you would find something that is completely orthogonal to what your mind is capable of understanding.
To me, a human mind conjuring up a working AI is akin to the human mind actually visualising higher dimensions: its too abstract.
We are flat landers in the world of AI, we'll never achieve it.
I've tried pointing this out to people before, but usually it falls on deaf ears: The left wing is the side of the political fence which is traditionally associated with censorship, control and social conformity.
I'm too lazy to dig it up now, but there was a quote floating around the place from the censorship board in which they confirmed that if a girl appears to be underage, the content is refused classification. So yes, they do actively censor women with small breasts.
Can you elaborate? That is not particularly scientific reasoning there.
It is known that on geological timescales, carbon dioxide was at much much higher atmospheric concentrations than it is today (here's an example), and yet global surface temperatures were not much different to what they are today at these times. I've still found nobody who can explain how this fits into current AGW dogma.
Water vapour is a potent greenhouse gas, anyone who has lived in a humid area knows this intuitively. CO2 is a very weak greenhouse gas, this is well known to scientists, the energy required to cause the 'predicted' temperature rises is understood to be far greater than what can be accounted for in CO2 alone. All predictions are based off a 'catalyst' assumption that minute increases in atmospheric energy stored in CO2 molecules will precipitate greater effects in the more potent greenhouse gases.
I'm not going to pass judgement on the latter part of this argument, but I will say this: Strong evidence exists to suggest that atmospheric CO2 is not an indicator of global surface temperature, it is also known that CO2 itself would never heat up the earth sufficiently to meet predictions. Between these two points I see it as somewhat dubious to attempt to link up the rest of the argument in favour of AGW predictions.
BigDog runs on a 2-stroke petrol engine, its limbs are actuated by hydraulics which are controlled by computer. All you gotta do is fill up its tank, no time wasting and infrastructure dependent recharge. Yet another way in which the ever denounced ICE is superior to all battery alternatives.
Slashdot Mirror
CS has actually been used in the seismic community since the 70s, seismologists have simply been assuming they could do this because thats all the data they had to work with.
In 2006, Terry Tao and Emmanuel Candes proved that this is true, and described the framework under which it is true.
From what I understand, neither of these statements are in fact the case. Firstly, Compressed sensing is capable of reconstructing data at significantly sub-Nyquist sampling rates, the Nyquist criterion does not apply at all in this case. Secondly, as far as I'm aware, JPG data will work fine, although I think it is possible that some noise may be interpreted as structure, but that is not to say it won't work. In fact that basis in which you construct the data is some sort of fourier, cosine or wavelet basis, which is pretty much what JPG does.
CS works by the assumption that your data is sparse on some basis, that completely rules out the idea that you can simply sample pixel data at random and reconstruct the image. What you have to do is sample data from a sparse basis, essentially you need to take random bits of information from something like the fourier transform or wavelet transform of the image, and what you can reconstruct is the complete set of information about this transformed data set. There is no magic tricky algorithm, just a very very sophisticated proof, in fact the algorithm for compressed sensing (there are many) is any linear programming algorithm (simplex, interior point, etc). Compressed sensing says that if you randomly sample data points from a sparse data set, then the true data set is the most sparse reconstruction you can make that agrees with your sample. This is done by minimising the l1 norm of your data set, given that your reconstruction remains consistent with your sample set. Or more precisely, given an unknown data set x, random sampling matrix A and the resultant sample, b, x can be recovered by the following:
min ||x||_1 such that Ax = b
The logic behind it is completely abstract from any notion that you need to know something about the image to reconstruct it (other than the assumption that it is sparse), you are not so much inferring details of the image from the sampled data as you are saying this: assuming that your data set is sparse, a random selection of that data set contains all the information about its sparsity.
They also, I believe, have the highest educated population in the world. You can't get to where Germany has gotten with a private-sector controlled education system either, so you can score two for Germany.
That said I have heard they have issues with people actually entering the workforce, since so many people stay on at university to complete a PhD, they all end up starting out in the workforce at around 30. Mind you it doesn't seem to be doing too much damage, although they have a smaller per capita gpd than Australia...
The only problem I see with it being adopted by the casual crowd is the fact that it needs to be tethered to a PC. Everyone keeps saying "oh I could easily give this to my mother", but then she still needs a PC with iTunes to run the thing right?
Thats not exactly how it works, the power "assists" your steering, there is always a mechanical linkage so that you can control the car, you just get a boost from the hydrualics. The thing is power steering allows for more favourable suspension geometries which result in heavy steering.
Its more or less the same story with your brakes.
Why is it so dire? Can you give me any credible source of research that says we face utter extinction if we do not stop the CO2 emissions? And what is it with the resources FUD? Where are they going? are we shooting all our resources out into space or something? Matter more or less never changes in any way other than configuration, our mass is going more or less nowhere.
There is a standard for_each, but I think its cumbersome in its own way, and there is a foreach in boost. Heres the boost example code for iterating over a 2d vector:
Concise, but still somewhat obtuse. The python equiv is still much clearer in its intent, imo. You can just iterate over multidimensional lists in Python and they will be flattened. You can even use Numpy and use slicing syntax to perform the same ends:
Its not even that the iterator line is ugly, its the fact that you have to create one in the first place, and its a pointer. I like coding in C++, the output is solid and you feel like you've really achieved something proper, but it is inherently more verbose, obscure, and unreadable than Python.
Python sacrifices a lot for readability, and it really is the most readable language out there.
That is true, but many languages carry with them a lot of housekeeping to get things done, Python tends to keep that sort of stuff out of the domain of the code. Consider a simple example, in C++ say we create a vector and iterate over its elements, we need to make the vector, an iterator, then invoke some pointer arithmetic to actually loop, and then dereference the iterator to extract data. But the important thing here is that we're iterating over items in a list (vector), however you end up with something like:
In Python this is simply:
You can set the C++ out as neatly as you want but the Python is inherently more readable because the housekeeping aspects are removed from the code. I don't care how much of a guru you are, to any human the Python example is infinitely more readable, and more of the code is devoted to describing exactly what is happening, rather than tangential intermediate steps. Yes this means you are trusting Python to do your housekeeping for you, and yes this means a performance hit, but this is besides the point, you shouldn't use Python in an setting where these things are important.
This is not to rag on C++, I really like C++, but it silly to say that Python, probably the most readable language ever invented, is no more readable than any other language. And don't get me started on Java, but then I don't like Java at all.
Exactly how much productive work have you done in Python? Because if you think anyone can output code faster in C than Python, you kind of sound like you've never used python.
"if you think python is the answer, you are kidding yourself."
You'd better tell Google that, maybe they'll start porting all their Python over to PHP?
"...everything else php does in a much more flexible fashion...."
"...it also allows geniuses to implement procedural processes much more elegantly."
Let me guess, you are 15?
I think you have the Prius figure a little off, all literature I've read, and accounts from people as well, put it fairly solidy in the 7L/100k range (in realistic terms). To put that into perspective, my 1992 carburettor powered honda civic gets about 7.5L/100k at its peak efficiency.
What MPG (or L/100km as we have down here) does say though is something about *efficiency* at the same time as energy consumption, because we can quite easily relate the notion of gallons (or litres) to our wallet. We can then determine very easily the economic efficiency of our vehicle. This is what is more important to people than outright energy consumption, and this is the only way you can change their mind about consumption habits as well.
Oh and an electric vehicle still consumes resources, so making remarks about infinite MPG is kind of stupid given they are essentially running on stored energy from coal fired steam turbines. So how many miles per unit of coal do you get then?
The government is not running on the assumption that a filter will save all the children, although it certainly is in their PR arsenal. What the government is doing, in their own eyes, is simply closing a loophole in the law. Australia has tough censorship laws *already*, the internet is just not filtered at this point. The government is simply seeking to apply its existing legal framework to the internet.
But this raises the far more important issue: Australia has a draconian censorship framework which needs to be brought into the modern age. The mere fact that the government applying their ratings rules is immediately seen as great wall of China style censorship is indicative of how out of touch the local censorship laws are with contemporary society. This sums up the far bigger problem that critics face: Conroy see this whole issue as applying the law in its intent, the way it is meant to be applied (closing a loophole), in order to get the government to view it any other way would require the government to be convinced it censorship framework needs to be loosened up, R18+ for games would be a nice start!!
Is that any reason to go looking to the state to enforce censorship on an entire country? So that you can have the freedom to not install an internet filter on your own computer?
Surely the OEMs would have put in massive preorders to have Win7 machines ready at launch?
No, as I understand it we know fully well that we are not a simple conglomeration of algorithms, isn't that the point of the halting problem? Or incompleteness for that matter (although that is somewhat more esoteric).
Yes a computer is nothing more than a series of algorithms, and this is precisely why it will never be an AI (please note I see an AI as not the mere illusion of intelligence, but true intelligence, created artificially). At least until we can get beyond Turing machines, but then you have to ask if we can even begin to grasp what it means to be intelligent in the first place. And I suspect we don't.
Is any of that actually intelligence though? I see it as sophisticated algorithms, modern advanced mathematics put into practice, and so forth. I don't see much intelligence out of any of the examples you gave and I honestly don't think those sorts of things are what the article was talking about. Autonomy is simply reacting to your environment with a predefined set of instructions. Intelligence is something far more than that.
I'm going to weigh in here and back the GP up. You don't need to stretch the imagination so far as to say this means we have a soul, Kurt Goedel essentially *proved* what this man is trying to say in a cold calculated and highly mathematical manner, back in the 20s. Its depends how you want to interpret the implications of his proof, the actual statement is something to the tune of this: there exists truths about natural numbers which are not provable under any axiomatic system. Fine, not really much to do with the human brain, but this idea is essentially what led Turing to go on a concretely prove that the Halting problem is undecidable. Again, seemingly innocuous? But the implications of these two gigantic theorems are straight forward: Firstly Turing showed without any possibility for argument that a Turing machine can *never* achieve intelligence, more importantly Goedels works shows that the human brain is capable of understanding truths for which we cannot explain systematically, he actually believed this was the core of human intuition.
So how do we proceed from there? We know that we cannot systematically describe everything which the human brain is capable of understanding, so how do you imitate its function if you can't even describe it in the first place?
I'm pretty sure you are right, and I'm pretty sure I strongly agree with you. Goedel showed that the human mind is capable of conceiving things beyond the capability of any formal system devised by the human mind, Turing showed that not Turing-complete machine can can solve all problems. Between these two giants and the vast body of work that has grown from their findings, I remain utterly skeptical that AI is even something humans can do. There is something eerily abstract about the whole notion.
I always love to pose this question to people on this topic and similar ones: How can you understand how your mind works from an outside perspective. You can't, its a logical fallacy, everything, absolutely every thought you have ever is the consequence of a human mind at work. You cannot escape that framework and objectively analyse it, and I suspect if you could, you would find something that is completely orthogonal to what your mind is capable of understanding.
To me, a human mind conjuring up a working AI is akin to the human mind actually visualising higher dimensions: its too abstract.
We are flat landers in the world of AI, we'll never achieve it.
I've tried pointing this out to people before, but usually it falls on deaf ears: The left wing is the side of the political fence which is traditionally associated with censorship, control and social conformity.
I'm too lazy to dig it up now, but there was a quote floating around the place from the censorship board in which they confirmed that if a girl appears to be underage, the content is refused classification. So yes, they do actively censor women with small breasts.
"There's backwards compatiblity for things that ran on it."
No there isnt, that went out the window with Win98/ME. NT needs NT stuff and the old DOS based stuff won't run on it.
Dude, if you 'tune' it differently (read: recompile with completely different sets of code) is it *really* the same kernel anymore?
"YES WE DO you FUCKING idiot."
Can you elaborate? That is not particularly scientific reasoning there.
It is known that on geological timescales, carbon dioxide was at much much higher atmospheric concentrations than it is today (here's an example), and yet global surface temperatures were not much different to what they are today at these times. I've still found nobody who can explain how this fits into current AGW dogma.
Water vapour is a potent greenhouse gas, anyone who has lived in a humid area knows this intuitively. CO2 is a very weak greenhouse gas, this is well known to scientists, the energy required to cause the 'predicted' temperature rises is understood to be far greater than what can be accounted for in CO2 alone. All predictions are based off a 'catalyst' assumption that minute increases in atmospheric energy stored in CO2 molecules will precipitate greater effects in the more potent greenhouse gases.
I'm not going to pass judgement on the latter part of this argument, but I will say this: Strong evidence exists to suggest that atmospheric CO2 is not an indicator of global surface temperature, it is also known that CO2 itself would never heat up the earth sufficiently to meet predictions. Between these two points I see it as somewhat dubious to attempt to link up the rest of the argument in favour of AGW predictions.
In short: Occam's Razor is not being applied.
"...requires recharging..."
BigDog runs on a 2-stroke petrol engine, its limbs are actuated by hydraulics which are controlled by computer. All you gotta do is fill up its tank, no time wasting and infrastructure dependent recharge. Yet another way in which the ever denounced ICE is superior to all battery alternatives.