A neural network (well, anything more complex than the
single-layer perceptron [wikipedia.org] anyway) is an arbitrary
classifier. I'm curious as to why other methods are "much
better". Unless you do an exhaustive search of the feature-space, all
classifier methods are subject to the same limitations - local
maxima/minima (depending on the algorithm), noise effects, and data
dependencies. All of the various algorithms have strengths and
weaknesses - in pattern recognition (my field) NN's are pretty darn
good actually.
Actually, more recent methods don't have local maxima/minima. Something
like a support vector machine optimizes an objective function. Of
course, this is somewhat of a tangent, in that the objective function
might not be a useful metric for performance, but people have shown that
the minimum objective function value of a SVM does relate to its generalization
performance. It's a little disconcerting that a NN has an objective
function but that it can find it's minimum or that the minimum doesn't
give good performance on test data (over-fitting)...
Of course, part of the NN's problem stems from the fact that it is
an arbitrary classifier. It's hard to give generalization results for
an algorithm that has an infinite VC dimension. (There are techniques
to restrict the size of the weights to give some guarantees.)
However, this doesn't mean NNs can't perform well in practice. It
probably means that the current theoretical analysis is somewhat
flawed in relation to the real world.
So have you ever compared your NN algorithms with the popular
algorithms of the day such as SVMs with kernels or boosting
algorithms. Also, are your NN algorithms generic or do you heavily
customize and tweak to get good performance.
One has to wonder who is doing the work. Is the paralyzed man adapting to the computer or is the computer learning the brain signals. Either way, it's good work, but I would bet that the way to perfect this type of technology is to "teach" the human to control his neurological impulses. I doubt the technology is directly eavesdropping on his speech.
The problem is that programs written in your normal
languages (C, C++, Java, C#, basically anything you've ever heard of)
are totally synchronous; you can not proceed on to the next statement
until the previous one completes.
Thus, trying to parallelize something at the API is virtually
worthless. I don't win anything if my "drawWindow" or
"displayMPEGFrame" function flies off to another processor to do its
work, if I still have to wait for it to complete before I can move
on.
The point of parallelizing the API, in this case, is not to execute
the API in parallel with the main program but to have the API execute
parallel code. This was the point of the Information Week article.
Of course, a much more sophisticated API is needed that targets a
flexible and general set of expensive parallelizable operations. The
goal would then be to write a sequential program that is expressed in
terms of the parallel API. This is suboptimal, but it does push the
difficulty of parallelization to the API.
It seems your main argument is the ends justify the means, and we
need to take the fight to the Islamic fundamentalists before they grow
too strong. If we don't, you predict that in just a couple of
generations an apocalyptic religious war will destroy our way of life.
Even if were to accept this (which I don't for so many reasons),
you don't explain how the war in Iraq is accomplishing this goal. In
fact, it seems to be doing the opposite. The US removed a secular
government and gave the fundamentalists an opportunity to turn Iraq
into an Islamic state. The US fuels hatred of Americans in the region
and is helping fundamentalists recruit a new generation of terrorists.
The US is putting Americans into a hostile region where they are easy
target for terrorists and insurgents. The US alienated much of the
world with this preemptive war. So based on your justification, this
war has been a massive failure.
However, I tend to agree with the GP. This war is about energy
dependence. The US wanted more influence in this region, and Iraq was
a convenient target. It was all spelled out years before 9/11 by the
neoconservatives. Of course, no matter what the justification, the
war has been a poorly planned failure.
I also agree with the GP's claim that a smarter strategy is to
spend more money researching alternative energy. If we developed
technology that removed the worlds dependence on oil then future
administrations would spend a lot less time dealing with the Middle
East. Many of those countries economies would collapse, and the
Islamic terrorists would have a hard time getting funded. Of course,
the Bush administration will only give lip service to alternative
energy. Hopefully someone in the future will do what's best for
national security.
I have to agree with the GP. Medicine, in many ways, seems very
primitive compared to other sciences. Despite it's tremendous
importance and the amount of money we spend on research, progress
seems very slow. The problem is that there is little consistent theory
to tie together the ideas. Unlike physics or chemistry, which have
depth, medicine looks very flat with lots of little discoveries that
are not obviously connected. Perhaps that's just the nature of the
beast, but it seems the human mind isn't good at comprehending all
these little facts that interact in unobvious ways.
As a physician I feel qualified to respond. Care to lend
parts of your body for experimentation? I can't promise you that
you'll survive. I can't promise that you won't be disfigured. And I
can't promise that you won't die from the consequences of some
unforeseen side-effect. No? I didn't think so somehow. We're bound by
ethics to try things only when we're almost completely sure they will
work and "do no harm".
While it would speed up things to work with humans, medicine has
always had animal models. While not identical to humans, they have
much of the same cellular machinery that we still don't understand...
I find it amusing how you can compare say coronary artery
bypass grafting, or a laparoscopic hernia reduction, with Egyptians
drilling holes in people's heads. They did it, yes. Now how many
people survived the procedure?
No doubt we have made progress, but there is so much that is still
a mystery. A large part of medicine is still based primarily on
experimentation. Create a bunch of chemicals and experiment to see if
any have interesting properties. Take those and experiment with them
on animals. If any of those work start experimenting on humans. Yes
rational drug discovery is making some headway, but things still seem
pretty primitive. Check out this
article to see how many people are killed by modern medicine.
As for the X rays and film, I believe I can introduce you
to the CT scanner, a device now so affordable that most hospitals have
several - even one _inside_ the ER. The film is still used for a hard
copy, but it's printed by computer. Oh speaking of X-rays, I suggest
you have a look at all the virtual endoscopy that's being done now,
with 3-D modelling software. I can see inside your blood vessels
without even touching your body. Let's not mention MRI's or PET scans
shall we? No X-rays involved there at all. Quite a bit of progress
since 1800. Radiology is one of the fields that is booming. Those
radiologists are going to put us all out of work, I tell
you.
While these are nice technical achievements they don't directly lead
to better understanding. Hopefully, the better images we get from
these tools will lead to achieve better understanding and not just
more empirical results. For example, the MRI research on brain
function doesn't seem to be leading to a better understanding of how
the brain works. However, maybe it's a important early step.
The most common method for curing infections? Actually
penicillin is hardly used nowadays, at least not at home. I invite you
to look into penicillin derived synthetics such as the cephalosporins,
aminopenicillins, ureidopenicillins. Then we have entire new classes
of antibiotics, from macrolides to fluoroquinolones to
aminoglucosides. Never heard of imipenem and meropenem? Most people
haven't. How about vancomycin, or linezolid for that matter? I just
named almost a dozen different families of antibiotics, each with
different biochemical mechanisms.
Well brawn (evolution) just might beat out brains. The number of drug
resistant bacteria are increasing. Doesn't seem like a success story.
Most democratic counties are not protected by such a take over because
it seems so unlikely. However they should be protected because
democracy can be abused. That is why governments should also
guarantee liberty. People need to be protected from the tyranny of
the majority. Read On Liberty by
John Stuart Mill to see why this is important. It's liberty that is
inconsistent with certain aspects of Sharia Law.
When it's got a looser planting on it, then it's
protected from neither. It ends up in rivers, where it washes out to
the coast and creates anaerobic conditions there; this kills sea life,
especially in harbors, river mouths, bays, et cetera. This also harms
the community of plants along the coast, and it's one of the reasons
why New Orleans got run over by the weather last year - the natural
windbreaks are pretty much nonexistent today.
This whole discussion is filled with so much misinformation...
With respect to New Orleans, the opposite is true. The flooding of
the Mississippi is what replenished the wetlands and protected New
Orleans from storm surge. The creation of the levees prevented the
river from flooding and ironically prevented new soil from maintaining
the wetlands. Check out
this for more information.
As for windbreaks, I doubt the wetlands have much effect, but any
effect is diminished as the wetlands shrink due to lack of
replenishment.
But the societal problems of alcohol use remained. Druken driving,
domestic abuse, chronic alcohol abuse, physical problems stemming from
chronic alcohol use, etc.
No one claims legalization will fix all problems, but it's still a
better world to fix some of the problems than none of the problems.
Legalization might even make some particular problems worse. However,
one needs to compare the total effects of both policies to make a
logical choice. In particular, many of the undesirable effects of
illegal drugs are really a side effect of them being illegal.
The problem is that simply legalizing dangerous drugs in
a complex society is fraught with lots of other problems. Yes, tiny
little countries in Europe have experimented with legalization and
government control of some very powerful, addicting drugs - I am not
sure that this model would translate well in the US. I am also not
sure of what mix of regulation and prohibition of drugs would be
appropriate in the US, but I am sure the answers are neither
simplistic nor easily attained.
Luckily people have already done lots of research. In some ways, it's a
cost benefit analysis. Of course, one of the biggest problems is that
people don't know and understand the issues well enough to do a
logical cost benefit analysis.
After doing research, I've come to the conclusion that legalization is
a better solution in terms of liberty, economics, and harm reduction.
Of course, legalization doesn't necessarily mean selling drugs out of
candy machines. The primary goal is to eliminate the black market and
there are many possible legalization strategies...
It's pretty easy to come up with an economic theory to explain why
unions achieve, over the long term, better compensation, but also a
bit pointless. The theory is supposed to explain and predict the
reality. Economics is a complicated system, and currently the theory
doesn't do a very good job. Therefore, if you want to answer your
question just look at reality.
Not quite true. The GPL does not (and cannot) reduce the
rights of the copyright owner. The only thing that can do that is
assignment of the copyright to another entity.
Well I guess it depends on how one defines rights. For example,
before licensing under the GPL, the copyright owner can stop anyone
from publishing his code. Afterwords, anyone who receives the code
from the owner can publish the code.
The GPL can not be unilaterally revoked. This means that
code, once licensed under the GPL, remains under the GPL. The
copyright owner is still free to release it under other licenses,
however. If they do, then users of the code may have the choice as to
which license they accept. An example of this is MySQL; you may either
use the GPL (and abide by its conditions) or buy a proprietary license
with fewer constraints.
Yes, this was my point. I guess it was confusing when I said the
owner can not change the license. By this I mean he can not change
the license to, for example, effectively revoke the GPL license. He
can still distribute his code under a different license. I assume
this is the "right" of the copyright owner to which you refer.
This is certainly true for the current version, but for
upcoming versions the copyright owner can change the license. So users
of GPL software does not have to worry that their software suddenly
becomes illegal.
But still, if I would refrain from contributing to projects that
requested copyright transfers, because they might take my code
proprietary in the future. If I wanted to allow that, I would use a
BSD (or similar) license instead.
I guess I'm not being clear because of the hour, but my point was
that under the current version your software cannot become illegal
even if you transfer the copyright. If the new owner releases a
proprietary version, the old version is still under the GPL, and you
can use and build on that old version.
Perhaps I misunderstood your original point, but the copyright
owner can only lock up the code if he has the copyright on all the
code and none of the code was ever released under the GPL. Therefore,
if you want to protect your contribution then you just need to release
your contributions under the GPL before you give up the copyright.
From then on your contribution will always be available.
In addition, you should only contribute to a software project with
a license that always guarantees you access to the code, since your
contribution might not be worth much without access to the rest of the
code. I assume this is part of what it means to be a GPL compatible
license.
However, maybe you just don't like the fact that someone can "lock
up" a version of your contribution and profit off your labor. In that
case, you should keep the copyright.
The is one advantage of this not yet mentioned in this
thread: By retaining the copyright to the code you wrote, you can
prevent Linus (or the suggested oversight body) changing the license
to a proprietary one, locking the code up along with your
contributions.
As understand the GPL license, even the copyright owner can not
change the license. If you could change the license then this would
be a big problem for anyone who depended on GPL software. At any time
your software could become illegal. Instead, what the copyright owner
can do is release his software under multiple licenses. Therefore, if
some owns the copyright they can effectively create a proprietary
fork but anyone can take up the old GPL software and continue
developing it.
I don't think you understood what the AC was saying. In today's
society, we need these minimum/low wage earners to keep things
running. That's how society is set up. Everybody can't train and get
better jobs because there are so many necessary low skill jobs.
So given that these low skill jobs exist, do you think it's OK for our
market economy to set their wages. Does the market take into account
cost of living and social responsibility? Do you care. Why do you
hate these people so much that you want to give them as little as
possible. As the AC said, these low skill workers are the ones who
actually do the work and keep this country running. You act as if
they are the ones screwing us over. They are the ones doing the
crappy low pay jobs that allow you to live your comfortable life.
Simple physics. The amount of energy it takes to destroy
an even small portion of the surface is incredible. I've played around
with the figures on a few antimatter devices. Even enough antimatter
to send a ship on a constant-acceleration course to Proxima Centauri
wouldn't be enough to more than scar the Earth. About the best you can
do is throw a very large asteroid at the Earth. This presents several
problems, though:...
First, simple physics arguments often turn out to be wrong because
of faulty assumptions. You assume some type of explosion is
necessary. There are many possible ways technology could destroy all
humans. I'm sure we'll invent many unexpected things in the next 100
years.
Second, even your argument based on simple physics isn't right.
You assume that it requires too much energy, but how do you bound the
amount of energy we can access. Who's to say that in 100 years
someone can't turn the entire earth into a bomb.
3. Being able to move a large asteroid implies space
travel. Which further implies a strong liklihood of off-Earth
colonies.
Space colonization is the answer Hawking is fishing for. It's
definitely the long term solution, but perhaps it's also the best bet
for the near future. It has the nice side effect that if the
scientists are spending all their energy on space colonization, maybe
they will be less likely to develop the next great weapon. (As I
implied before, even if this weapon doesn't kill all humans, just
destroying civilization will significantly decrease the expected
lifetime of humanity since it will make space colonization much more
difficult.)
Our big shiny toys put unprecedented powers of destruct\ion in the
hands of very few people. We are quite capable of destroying ourselves
now in a way that would not have been easy in the past.
Two points:
1. It would take a weapon larger than anything we have in existence to kill \off the human race. Cultural segregation or no, at least some people will survi\ve even the most brutal attempts of destruction. As it so happens, all of our "\weapons of mass destruction" combined are barely enough to wipe out all the maj\or population centers.
So I take it you never believed the saying that we had a enough
nuclear weapons to destroy the world several times over. Also you
seem to doubt the ingenuity of humans. Don't you think we will
develop more powerful weapons over time? What's to stop one of these
future weapons from being a global killer.
(Ironically, we might be safer from ourselves if we destroyed modern
civilization with our modern weapons. That way it would be
significantly harder to develop more powerful weapons. Of course then
we would never get off this planet, which is a concern for the long
term survivability of the human race.)
2. Scale the matter back to local civilizations. Come
up with an example or two where a civilization managed to completely
destroy itself. I can't really come up with any good examples myself,
though I can come up with a few examples of civilizations destroyed by
other civilizations. (But not completely! The survivors are almost
always integrated into the conquering civilization.) As I said, our
instinct for survival is too high.
The point of his post was that history is not a good predictor since
technology makes destruction easier for individuals and technology
keeps advancing.
But no one has the right to accumulate masses of wealth without
consideration for others
Actually, everyone has that right. It's fundamental to
economic freedom.
Not every conceivable freedom is a good thing, and every right has
limitations. In this case, I doubt many people have this aspect of
economic freedom since all governments have an involuntary tax of one
form or another.
Your benchmark is pointless. I just wrote a quick program to calculate 1,000,000 factorial (ten times your number). I ran it on an AMD 800 thunderbird. It took 11 seconds. Blam - my 5 year old AMD is nearly 100 times faster than your modern P4 3.2GHz.
My program is even faster. It just returns Inf. It's almost infinitely faster.
This really depends on where you get your numbers from:-)
Of course, I was giving a rough justification. I encourage people
to read the references I gave for more details. But you do have a
point, making additional assumptions about where the numbers come from
and what operations you are going to perform can often be used to
greatly reduce round-off error. However, in general, if you are doing
floating point rounding with numbers that are already rounded then
round to even is preferred.
Consider numbers actually consisting of three parts. The integral part
to where you wana round, the known fractional part that you base your
rounding on and a third part of following unknown digits. The last
part is usually consisting of all zeroes for monetary values. So your
number is:
xxx.yyzzz....
So in your example 1.0 is actually 1.0zzz with z representing unknown
digits. Which means that rounding 1.0 to 1 is NOT exact when talking
about natural numbers. Same goes for 1.5zzz, where rounding up may be
more precise if those z are nonzero.
As I already said, this means the numbers as given have been
truncated. However, as you do further calculation with these numbers
they will be rounded and therefore, rounding to even is preferred.
Ideally you want to start with rounded numbers since the initial
truncation will cause a lot of bias, but sometimes you can't control
that.
Bankers rounding is cute and nice for quite a few applications, but it
*can* also introduce a bias of its own, depending on what you do. I
hope you see now, why MarkusQ has a point.
I don't mean to knock MarkusQ, but he was giving bad information
and based on his moderation points people were believing him. I had
moderation points, but I figured it would be more informative to
explain why he was wrong. My hat is off to him. It's much easier to
explain a true fact than to convince people of a mistaken belief. (Or
at least I hope so.)
Re:Only with money in fractions
on
Rounding Algorithms
·
· Score: 2, Informative
The source you cite (and, it appears, your arguments, at least
implicitly) are talking about how to deal with repeated rounding in
calculations with a single guard digit. As the Knuth paper clearly
states in the introduction to the defense of round-ties-to-even:
Incorrect. Did you read the article? The rest of your quote from the
article is
The example immediately preceding Theorem 2 shows that a single
guard digit will not always give exactly rounded results. The previous
section gave several examples of algorithms that require a guard digit
in order to work properly. This section gives examples of algorithms
that require exact rounding.
so it has nothing to do with guard digits. In fact, they then go
on to define rounding, since at this point it hasn't been formally
defined. After they give the rounding to even definition, they
justify the definition with a theorem from Reiser and Knuth. Note,
the webpage is not a paper from Knuth; it is from David Goldberg.
Goldberg then gives an example to show how rounding up can introduce
bias in a way that rounding to even can not. Last they consider the
issue closed and assume rounding to even for the rest of the paper.
On the more general case, the only mention corresponds to my
position:
...double-rounding only produces an incorrectly rounded result
when the second rounding is determined by the round-ties-to-even
rule...
This is not a more general case. They are talking about a precise
type of problem called double-rounding that comes from mixing double
precision and extended double precision calculations. Furthermore
they don't talk about rounding up as a solution to this problem.
It's clear you just did some simple searches of the article to try
and support your position. I suggest you read some of the article.
Notice it's title, "What Every Computer Scientist Should Know About
Floating-Point Arithmetic".
Re:Only with money in fractions
on
Rounding Algorithms
·
· Score: 2, Insightful
For example, the number 4.56106531 would be rounded to 4 in the
"nearest even" case or to 5 in the "round up" case But clearly, the
"nearest even" result is less accurate, and introduces a significant
bias. 4.56106531 is closer to 5 than to 4, and should be rounded
up. Always.
This is the wrong rule. The number 4.56106531 would be rounded to 5
with both techniques. The "nearest even" technique for rounding to
the nearest integer only applies to 4.50000000. In this case, you
would round the number to 4. You might say what's the point, since
this is a very unlikely case. However, if you only have two digits to
work with it becomes much more significant.
At this point, you may object that you aren't planning on
truncating before you apply the rule (or, equivalently, you only do
the even odd dance on "exact" halves). But how did you get an "exact"
half? Unless you have infinite precision floating point hardware, less
significant bits fell off the bottom of your number; unless they were
all zero your "exact half" is the result of truncation and the above
logic still applies.
The "exact half", for all you know might be below.5 or might be
above.5. You seem to be assuming it is the result of truncating a
single number. The rounding occurs during arithmetic operations and,
for all you know, the real extra is just as likely to be below.5 as
it is above.5. That's the essence of rounding and why we don't use
truncation.
But just for fun, lets take the case where you are inputing the
number into the computer. If the number has a number of significant
digits more than the machine then for most problems the number is just
as likely to be a little below the given number as it is above,
therefore you should use "nearest even" to remove bias. If the number
has a number of significant digits less than or equal to the machine,
you just enter the number.
As an rough argument why "nearest even" works, every time you do an
arithmetic operation with rounded numbers the new number is just as
likely to be a little too big as it is to be a little too small.
(This is why we represent rounded number as x*(1 + or - epsilon).)
Therefore rounding 1.0 to 1 is ideal since the number is just a likely
to be a little smaller as it is to being a little bigger. Rounding
1.1 to 1 starts introducing bias since you are shrinking the
number. However, the bias for 1.1 balances out the bias for 1.9, and
the bias for 1.2 balances out the bias for 1.8. However, if you
continue, the bias for 1.5 is not balanced out and the sum of numbers
tends to grow. One way around this is to flip a coin for 1.5 and
round up half the time and down the other half. This is expensive, so
a cheap alternative is to just use the next digit as the random coin;
round up if it's odd and round down if it's even..
Last, as an appeal to experts, Reiser and Knuth [1975] disagree with
you. In fact, Knuth has a write up in one of his Art of Programming
books on the issue. A short but online explanation can be found here
in the section Exactly Rounded Operations.
One idea is to have a standardized open contract that is publicized by
the research and computer communitiy. This contract stipulates rules
that include, at a minimum, that the results of the funded research
will be published no matter what the results of the experiments. When
a study comes out, it can be used as a seal of approval to show that
the research was not siphoned out of a pool of negative results.
The real problem is to get the companies to use the contract.
However, there must be some companies that have clearly superior
products who would be confident enough to use the contract. They can
start a trend where statistical research without the contract is
considered tainted.
While many details need to be filled out, it would be a big
improvement over the current situation. Currently a wealthy company
can poison the statistics of research by fishing for good experiments.
This is a serious problem and calls into question all of your
published vendor supported research.
This is not about predicting the stock market. It is about using the idea of the stock market to make predictions.
The idea is to create a market and use the prices to make predictions. This is not new. In fact, the US government got some PR trouble when someone proposed to use this idea to predict terrorist attacks. Look at this workshop to get some pointers for further information.
Slashdot Mirror
Actually, more recent methods don't have local maxima/minima. Something like a support vector machine optimizes an objective function. Of course, this is somewhat of a tangent, in that the objective function might not be a useful metric for performance, but people have shown that the minimum objective function value of a SVM does relate to its generalization performance. It's a little disconcerting that a NN has an objective function but that it can find it's minimum or that the minimum doesn't give good performance on test data (over-fitting)...
Of course, part of the NN's problem stems from the fact that it is an arbitrary classifier. It's hard to give generalization results for an algorithm that has an infinite VC dimension. (There are techniques to restrict the size of the weights to give some guarantees.) However, this doesn't mean NNs can't perform well in practice. It probably means that the current theoretical analysis is somewhat flawed in relation to the real world.
So have you ever compared your NN algorithms with the popular algorithms of the day such as SVMs with kernels or boosting algorithms. Also, are your NN algorithms generic or do you heavily customize and tweak to get good performance.
One has to wonder who is doing the work. Is the paralyzed man adapting to the computer or is the computer learning the brain signals. Either way, it's good work, but I would bet that the way to perfect this type of technology is to "teach" the human to control his neurological impulses. I doubt the technology is directly eavesdropping on his speech.
One of the best posts I've read in a while. Spot on; I wish I had mod points.
The point of parallelizing the API, in this case, is not to execute the API in parallel with the main program but to have the API execute parallel code. This was the point of the Information Week article. Of course, a much more sophisticated API is needed that targets a flexible and general set of expensive parallelizable operations. The goal would then be to write a sequential program that is expressed in terms of the parallel API. This is suboptimal, but it does push the difficulty of parallelization to the API.
It seems your main argument is the ends justify the means, and we need to take the fight to the Islamic fundamentalists before they grow too strong. If we don't, you predict that in just a couple of generations an apocalyptic religious war will destroy our way of life.
Even if were to accept this (which I don't for so many reasons), you don't explain how the war in Iraq is accomplishing this goal. In fact, it seems to be doing the opposite. The US removed a secular government and gave the fundamentalists an opportunity to turn Iraq into an Islamic state. The US fuels hatred of Americans in the region and is helping fundamentalists recruit a new generation of terrorists. The US is putting Americans into a hostile region where they are easy target for terrorists and insurgents. The US alienated much of the world with this preemptive war. So based on your justification, this war has been a massive failure.
However, I tend to agree with the GP. This war is about energy dependence. The US wanted more influence in this region, and Iraq was a convenient target. It was all spelled out years before 9/11 by the neoconservatives. Of course, no matter what the justification, the war has been a poorly planned failure.
I also agree with the GP's claim that a smarter strategy is to spend more money researching alternative energy. If we developed technology that removed the worlds dependence on oil then future administrations would spend a lot less time dealing with the Middle East. Many of those countries economies would collapse, and the Islamic terrorists would have a hard time getting funded. Of course, the Bush administration will only give lip service to alternative energy. Hopefully someone in the future will do what's best for national security.
I have to agree with the GP. Medicine, in many ways, seems very primitive compared to other sciences. Despite it's tremendous importance and the amount of money we spend on research, progress seems very slow. The problem is that there is little consistent theory to tie together the ideas. Unlike physics or chemistry, which have depth, medicine looks very flat with lots of little discoveries that are not obviously connected. Perhaps that's just the nature of the beast, but it seems the human mind isn't good at comprehending all these little facts that interact in unobvious ways.
While it would speed up things to work with humans, medicine has always had animal models. While not identical to humans, they have much of the same cellular machinery that we still don't understand...
No doubt we have made progress, but there is so much that is still a mystery. A large part of medicine is still based primarily on experimentation. Create a bunch of chemicals and experiment to see if any have interesting properties. Take those and experiment with them on animals. If any of those work start experimenting on humans. Yes rational drug discovery is making some headway, but things still seem pretty primitive. Check out this article to see how many people are killed by modern medicine.
While these are nice technical achievements they don't directly lead to better understanding. Hopefully, the better images we get from these tools will lead to achieve better understanding and not just more empirical results. For example, the MRI research on brain function doesn't seem to be leading to a better understanding of how the brain works. However, maybe it's a important early step.
Well brawn (evolution) just might beat out brains. The number of drug resistant bacteria are increasing. Doesn't seem like a success story.
So you work for the CIA?
Most democratic counties are not protected by such a take over because it seems so unlikely. However they should be protected because democracy can be abused. That is why governments should also guarantee liberty. People need to be protected from the tyranny of the majority. Read On Liberty by John Stuart Mill to see why this is important. It's liberty that is inconsistent with certain aspects of Sharia Law.
This whole discussion is filled with so much misinformation...
With respect to New Orleans, the opposite is true. The flooding of the Mississippi is what replenished the wetlands and protected New Orleans from storm surge. The creation of the levees prevented the river from flooding and ironically prevented new soil from maintaining the wetlands. Check out this for more information.
As for windbreaks, I doubt the wetlands have much effect, but any effect is diminished as the wetlands shrink due to lack of replenishment.
No one claims legalization will fix all problems, but it's still a better world to fix some of the problems than none of the problems. Legalization might even make some particular problems worse. However, one needs to compare the total effects of both policies to make a logical choice. In particular, many of the undesirable effects of illegal drugs are really a side effect of them being illegal.
Luckily people have already done lots of research. In some ways, it's a cost benefit analysis. Of course, one of the biggest problems is that people don't know and understand the issues well enough to do a logical cost benefit analysis.
After doing research, I've come to the conclusion that legalization is a better solution in terms of liberty, economics, and harm reduction. Of course, legalization doesn't necessarily mean selling drugs out of candy machines. The primary goal is to eliminate the black market and there are many possible legalization strategies...
It's pretty easy to come up with an economic theory to explain why unions achieve, over the long term, better compensation, but also a bit pointless. The theory is supposed to explain and predict the reality. Economics is a complicated system, and currently the theory doesn't do a very good job. Therefore, if you want to answer your question just look at reality.
Well I guess it depends on how one defines rights. For example, before licensing under the GPL, the copyright owner can stop anyone from publishing his code. Afterwords, anyone who receives the code from the owner can publish the code.
Yes, this was my point. I guess it was confusing when I said the owner can not change the license. By this I mean he can not change the license to, for example, effectively revoke the GPL license. He can still distribute his code under a different license. I assume this is the "right" of the copyright owner to which you refer.
I guess I'm not being clear because of the hour, but my point was that under the current version your software cannot become illegal even if you transfer the copyright. If the new owner releases a proprietary version, the old version is still under the GPL, and you can use and build on that old version.
Perhaps I misunderstood your original point, but the copyright owner can only lock up the code if he has the copyright on all the code and none of the code was ever released under the GPL. Therefore, if you want to protect your contribution then you just need to release your contributions under the GPL before you give up the copyright. From then on your contribution will always be available.
In addition, you should only contribute to a software project with a license that always guarantees you access to the code, since your contribution might not be worth much without access to the rest of the code. I assume this is part of what it means to be a GPL compatible license.
However, maybe you just don't like the fact that someone can "lock up" a version of your contribution and profit off your labor. In that case, you should keep the copyright.
As understand the GPL license, even the copyright owner can not change the license. If you could change the license then this would be a big problem for anyone who depended on GPL software. At any time your software could become illegal. Instead, what the copyright owner can do is release his software under multiple licenses. Therefore, if some owns the copyright they can effectively create a proprietary fork but anyone can take up the old GPL software and continue developing it.
I don't think you understood what the AC was saying. In today's society, we need these minimum/low wage earners to keep things running. That's how society is set up. Everybody can't train and get better jobs because there are so many necessary low skill jobs.
So given that these low skill jobs exist, do you think it's OK for our market economy to set their wages. Does the market take into account cost of living and social responsibility? Do you care. Why do you hate these people so much that you want to give them as little as possible. As the AC said, these low skill workers are the ones who actually do the work and keep this country running. You act as if they are the ones screwing us over. They are the ones doing the crappy low pay jobs that allow you to live your comfortable life.
First, simple physics arguments often turn out to be wrong because of faulty assumptions. You assume some type of explosion is necessary. There are many possible ways technology could destroy all humans. I'm sure we'll invent many unexpected things in the next 100 years.
Second, even your argument based on simple physics isn't right. You assume that it requires too much energy, but how do you bound the amount of energy we can access. Who's to say that in 100 years someone can't turn the entire earth into a bomb.
Space colonization is the answer Hawking is fishing for. It's definitely the long term solution, but perhaps it's also the best bet for the near future. It has the nice side effect that if the scientists are spending all their energy on space colonization, maybe they will be less likely to develop the next great weapon. (As I implied before, even if this weapon doesn't kill all humans, just destroying civilization will significantly decrease the expected lifetime of humanity since it will make space colonization much more difficult.)
So I take it you never believed the saying that we had a enough nuclear weapons to destroy the world several times over. Also you seem to doubt the ingenuity of humans. Don't you think we will develop more powerful weapons over time? What's to stop one of these future weapons from being a global killer. (Ironically, we might be safer from ourselves if we destroyed modern civilization with our modern weapons. That way it would be significantly harder to develop more powerful weapons. Of course then we would never get off this planet, which is a concern for the long term survivability of the human race.)
The point of his post was that history is not a good predictor since technology makes destruction easier for individuals and technology keeps advancing.Not every conceivable freedom is a good thing, and every right has limitations. In this case, I doubt many people have this aspect of economic freedom since all governments have an involuntary tax of one form or another.
Your benchmark is pointless. I just wrote a quick program to calculate 1,000,000 factorial (ten times your number). I ran it on an AMD 800 thunderbird. It took 11 seconds. Blam - my 5 year old AMD is nearly 100 times faster than your modern P4 3.2GHz.
My program is even faster. It just returns Inf. It's almost infinitely faster.
This really depends on where you get your numbers from:-)
Of course, I was giving a rough justification. I encourage people to read the references I gave for more details. But you do have a point, making additional assumptions about where the numbers come from and what operations you are going to perform can often be used to greatly reduce round-off error. However, in general, if you are doing floating point rounding with numbers that are already rounded then round to even is preferred.
Consider numbers actually consisting of three parts. The integral part to where you wana round, the known fractional part that you base your rounding on and a third part of following unknown digits. The last part is usually consisting of all zeroes for monetary values. So your number is:
xxx.yyzzz....
So in your example 1.0 is actually 1.0zzz with z representing unknown digits. Which means that rounding 1.0 to 1 is NOT exact when talking about natural numbers. Same goes for 1.5zzz, where rounding up may be more precise if those z are nonzero.
As I already said, this means the numbers as given have been truncated. However, as you do further calculation with these numbers they will be rounded and therefore, rounding to even is preferred. Ideally you want to start with rounded numbers since the initial truncation will cause a lot of bias, but sometimes you can't control that.
Bankers rounding is cute and nice for quite a few applications, but it *can* also introduce a bias of its own, depending on what you do. I hope you see now, why MarkusQ has a point.
I don't mean to knock MarkusQ, but he was giving bad information and based on his moderation points people were believing him. I had moderation points, but I figured it would be more informative to explain why he was wrong. My hat is off to him. It's much easier to explain a true fact than to convince people of a mistaken belief. (Or at least I hope so.)
For example, the number 4.56106531 would be rounded to 4 in the "nearest even" case or to 5 in the "round up" case But clearly, the "nearest even" result is less accurate, and introduces a significant bias. 4.56106531 is closer to 5 than to 4, and should be rounded up. Always.
This is the wrong rule. The number 4.56106531 would be rounded to 5 with both techniques. The "nearest even" technique for rounding to the nearest integer only applies to 4.50000000. In this case, you would round the number to 4. You might say what's the point, since this is a very unlikely case. However, if you only have two digits to work with it becomes much more significant.
At this point, you may object that you aren't planning on truncating before you apply the rule (or, equivalently, you only do the even odd dance on "exact" halves). But how did you get an "exact" half? Unless you have infinite precision floating point hardware, less significant bits fell off the bottom of your number; unless they were all zero your "exact half" is the result of truncation and the above logic still applies.
The "exact half", for all you know might be below .5 or might be
above .5. You seem to be assuming it is the result of truncating a
single number. The rounding occurs during arithmetic operations and,
for all you know, the real extra is just as likely to be below .5 as
it is above .5. That's the essence of rounding and why we don't use
truncation.
But just for fun, lets take the case where you are inputing the number into the computer. If the number has a number of significant digits more than the machine then for most problems the number is just as likely to be a little below the given number as it is above, therefore you should use "nearest even" to remove bias. If the number has a number of significant digits less than or equal to the machine, you just enter the number.
As an rough argument why "nearest even" works, every time you do an arithmetic operation with rounded numbers the new number is just as likely to be a little too big as it is to be a little too small. (This is why we represent rounded number as x*(1 + or - epsilon).) Therefore rounding 1.0 to 1 is ideal since the number is just a likely to be a little smaller as it is to being a little bigger. Rounding 1.1 to 1 starts introducing bias since you are shrinking the number. However, the bias for 1.1 balances out the bias for 1.9, and the bias for 1.2 balances out the bias for 1.8. However, if you continue, the bias for 1.5 is not balanced out and the sum of numbers tends to grow. One way around this is to flip a coin for 1.5 and round up half the time and down the other half. This is expensive, so a cheap alternative is to just use the next digit as the random coin; round up if it's odd and round down if it's even..
Last, as an appeal to experts, Reiser and Knuth [1975] disagree with you. In fact, Knuth has a write up in one of his Art of Programming books on the issue. A short but online explanation can be found here in the section Exactly Rounded Operations.
I didn't know Ed Begley Jr. posted on slashdot.
One idea is to have a standardized open contract that is publicized by the research and computer communitiy. This contract stipulates rules that include, at a minimum, that the results of the funded research will be published no matter what the results of the experiments. When a study comes out, it can be used as a seal of approval to show that the research was not siphoned out of a pool of negative results.
The real problem is to get the companies to use the contract. However, there must be some companies that have clearly superior products who would be confident enough to use the contract. They can start a trend where statistical research without the contract is considered tainted.
While many details need to be filled out, it would be a big improvement over the current situation. Currently a wealthy company can poison the statistics of research by fishing for good experiments. This is a serious problem and calls into question all of your published vendor supported research.
This is not about predicting the stock market. It is about using the idea of the stock market to make predictions.
The idea is to create a market and use the prices to make predictions. This is not new. In fact, the US government got some PR trouble when someone proposed to use this idea to predict terrorist attacks. Look at this workshop to get some pointers for further information.