Admins should create a system which counts how many same words stories have and AUTOMATICALLY show headlines of similar already published stories to the people how make decisision what to publish.
Or even better. Automatically make it impossible to publish stories which are too similar (use some simple count number of same words thing).
Anyone can call himself what he want. UK may disagree and not reocognize/use such title but it doesn't matter if people do.
(Lord British for example).
You are probably right but real coders code code which already uses the best possible O(f(n)) algorithm (maybe further improved/developed by himself). For serious computing the best algorithm can still be too slow.
I have problems/methods which can take hours/days/months to run (even when simplified to get crude answers): machine learning, matrix decomposition, monte carlo methods, fourier&wavelet transforms, principal&independent component analysis, variational bayesian methods, thousands or millions of variables.
I think headline is fair but the text should have mentioned $5 limit. "Microsoft Says Your Data Is Only Worth $5" is quite anti-microsoft.
Most companies pay you nothing. $5 x many million users is quite much already. If a serious bug(s) are found this means MS must pay of hundreds of millions.
"When I multiply numbers together, I see two shapes. The image starts to change and evolve, and a third shape emerges. That's the answer. It's mental imagery. It's like maths without having to think."
Mathematically speaking it seems like autists are able to create isomorphism between numbers and real world objects. People are very good at remembering and understanding properties, relations and interactions between physical objects.
I thought much about computer based isomorphisms between problem domains. Ideally people's problem solving skills could be used to solve problems in other domains if computer could for example transform math problems (and their solutions) to music composition problems etc. This could also make it possible to see problem from totally new perspective.
Because autists can find isomorphisms between different problems it then gives evidence that idea of trying to find isomorphisms between complex problem domains which are very different isn't impossible.
--
Tomas Ukkonen
If I remember correctly it has been shown that in certain special cases entropy of a closed system can decrease. I can't remember details but it was a very small system + maybe some other restrictions.
If anyone knows what I'm talking about please post link to the paper itself. I googled and find this
one which might the result I'm talking about.
Why would otherwise there would be concepts like sex, woman and man.
I believe (but don't have any solid research to refer) that because men and women hava had different roles during the evolution there might exist small differences (which may not be significant).
Men have been hunters so their 3d visualization and planning skills (and therefore math skills) may be better and because women have taken care of children their social skills are likely to be better.
> This is wrong as far as I'm concerned, just as claiming ownership over
> applied discrete mathematics == software is just wrong.
Patents doesn't mean you own something. You have only given a *temporarily* exclusive right to use a method.
Non-trivial software is like non-trivial physics. Just like physics isn't only an applied mathematical theory. Also software isn't only discrete mathematics. In both cases mathematics is only used as a language to describe nature/algorithms.
I'm currently against patents in general because standards used for issuing patents should be much higher. BUT if the standards can be rised (+ changes to couple of other things) then I think software patents are ok.
Non-patentability of software (europe) and mathematics is really slowing progress because the resulting IP (from costly R&D) cannot be protected.
This is not really that amazing. I had a similar idea a year ago but never had time to test/implement it.
This is also not a new idea. Someone wrote a program which optimized gcc's flags to give the best possible combination of optimization flags.
In general the use of optimization (statistical inference methods usually work better) allows adaption of programs and algorithms to hardware.
I would be more impressed if if systems could actively develop/learn (adation in a much more general sense and with 'memory') and plan ahead.
Example: software which learns which combinations of parameters cause function to crash (causal inference), has (or builds) a model how actions (instructions, interrupts) change system (learning bayesian networks + modelling by hand) and can use models & gathered information to workaround or fix (simple/some) bugs (planning, logic or statistical 'rule' models (I'm working on (hobby) [implementation&theory] these ones at the moment)).
This idea is really not that new. Try googling 'self-healing systems'.
Want some fame? Then just make above idea work in practice and release it OSS.
This is much more complicated and you must have studied AI, statistics and machine learning to be able to understand how to implement above.
Understanding (basic of) GAs is easy and so is implementation. They also work quite well. That's why they are so popular.
IMHO, if GA implementation can be made really reliable it could maybe replace other code which may be (I don't know) even more complicated.
Re:don't forget about darwinist programming
on
Subatomic Darwinism
·
· Score: 1
Here's a better answer.
(English isn't my native language so there are errors. I don't care).
Bayesian methods aren't same as GAs. I think GAs are crude, non-parametric approximations of bayesian inference.
GAs usually use very heuristical goodness and offspring creation and selection methods.
If they solve your problems that's obviously good enough for you but I think bayesian methods are better because:
Bayesian inference (and statistics) has much more sound theoretical backing in general and you know exacly what your assumptions about problem are
Because statistics is so well established you can usually choose distributions you use wisely so that they work well with your problem. There are also much research about non-informative prior distributions (as assumption-free as possible) which can be also used.
With bayesian inference you can put quite easily your prior information about solution into your model (so that solution can be found faster/easier). With GAs this is harder because you don't know exactly how your changes to parameters, offspring creation and selection changes your assumptions about correct solution.
There are statistical sampling methods that can be proved to scale well to high dimensional problems. AFAIK, books about GAs don't usually give such proofs, maybe because they don't exist?
By parameterizing distributions your optimization process is faster. (only simple cases can be parametrized well (conjugate distributions)). This is because now you don't have to have N individually competing genes.
As long as you can model distributions of data well, bayesian inference/statistics works well with any form of data (continuous, discrete). GAs only handle well bits and integers.
When using bayesian inferece you know what you are doing (if you have studied lots of statistics). With GAs you don't and you are only guessing. GAs are good approximations of bayesian inference but one should concentrate on baysian inference instead. Any far reaching research on GAs probably only duplicate efforts usually already made by statisticians.
Re:don't forget about darwinist programming
on
Subatomic Darwinism
·
· Score: 1
Why bother with GAs. They just special cases of baysian inference.
Say you want to find out probable good values for x (optimization, something else).
Define p(x) (starting population in GAs)
Define p(d|x) (related to goodness function of GAs)
sample from p(x|d)~ p(d|x)*p(x) (natural selection).
this is information theoretically optimal
(H(X|DATA) = H(X) - I(X,DATA) so it isn't surprise that natural selection does something very close to this.
Re:don't forget about darwinist programming
on
Subatomic Darwinism
·
· Score: 2, Insightful
Why bother with GAs. They just special cases of baysian inference.
Say you want to find out probable good values for x (optimization, something else).
Define p(x) (starting population in GAs)
Define p(d|x) (related to goodness function of GAs)
sample from p(x|d)~ p(d|x)*p(x) (natural selection).
this is information theoretically optimal
(H(X|DATA) = H(X) - I(X,DATA) so it isn't surprise that natural selection does something very close to this.
Yes. It is also predicted that US dollar will drop -30% within year. Now ask yourself if it would be smart move to exchange all your savings to EUROs, wait 1 year and get 30% profit.
Chances are that even if you can remove the causes of ageing which can be notices when people are 200 years old, there may be other ageing effects which don't show up until you are more than 500 years old.
Slashdot Mirror
Admins should create a system which counts how many same words stories have and AUTOMATICALLY show headlines of similar already published stories to the people how make decisision what to publish.
Or even better. Automatically make it impossible to publish stories which are too similar (use some simple count number of same words thing).
Links
Anyone can call himself what he want. UK may disagree and not reocognize/use such title but it doesn't matter if people do. (Lord British for example).
I have problems/methods which can take hours/days/months to run (even when simplified to get crude answers): machine learning, matrix decomposition, monte carlo methods, fourier&wavelet transforms, principal&independent component analysis, variational bayesian methods, thousands or millions of variables.
Compiler cannot reorder or do other optimizations with floating point code without potentially breaking something.
Most companies pay you nothing. $5 x many million users is quite much already. If a serious bug(s) are found this means MS must pay of hundreds of millions.
"When I multiply numbers together, I see two shapes. The image starts to change and evolve, and a third shape emerges. That's the answer. It's mental imagery. It's like maths without having to think."
Mathematically speaking it seems like autists are able to create isomorphism between numbers and real world objects. People are very good at remembering and understanding properties, relations and interactions between physical objects.
I thought much about computer based isomorphisms between problem domains. Ideally people's problem solving skills could be used to solve problems in other domains if computer could for example transform math problems (and their solutions) to music composition problems etc. This could also make it possible to see problem from totally new perspective.
Because autists can find isomorphisms between different problems it then gives evidence that idea of trying to find isomorphisms between complex problem domains which are very different isn't impossible. -- Tomas Ukkonen
Work done is also virtual. Music requires real skill s to create. MMRPG admins could create unlimited number of items if they wanted.
link
Me neither but Linux encryption (+ softraid) is really easy to setup if you have little experience.
If anyone knows what I'm talking about please post link to the paper itself.
I googled and find this one which might the result I'm talking about.
I believe (but don't have any solid research to refer) that because men and women hava had different roles during the evolution there might exist small differences (which may not be significant).
Men have been hunters so their 3d visualization and planning skills (and therefore math skills) may be better and because women have taken care of children their social skills are likely to be better.
Isn't that useful if you only instantate print_info() with handful of types.
> applied discrete mathematics == software is just wrong.
I'm currently against patents in general because standards used for issuing patents should be much higher. BUT if the standards can be rised (+ changes to couple of other things) then I think software patents are ok.
Non-patentability of software (europe) and mathematics is really slowing progress because the resulting IP (from costly R&D) cannot be protected.
Earth's environment were quite extreme then so any long-lasting molecules couldn't have been fragile.
This is also not a new idea. Someone wrote a program which optimized gcc's flags to give the best possible combination of optimization flags.
In general the use of optimization (statistical inference methods usually work better) allows adaption of programs and algorithms to hardware.
I would be more impressed if if systems could actively develop/learn (adation in a much more general sense and with 'memory') and plan ahead.
Example: software which learns which combinations of parameters cause function to crash (causal inference), has (or builds) a model how actions (instructions, interrupts) change system (learning bayesian networks + modelling by hand) and can use models & gathered information to workaround or fix (simple/some) bugs (planning, logic or statistical 'rule' models (I'm working on (hobby) [implementation&theory] these ones at the moment)).
This idea is really not that new. Try googling 'self-healing systems'.
Want some fame? Then just make above idea work in practice and release it OSS.
This is much more complicated and you must have studied AI, statistics and machine learning to be able to understand how to implement above.
IMHO, if GA implementation can be made really reliable it could maybe replace other code which may be (I don't know) even more complicated.
Bayesian methods aren't same as GAs. I think GAs are crude, non-parametric approximations of bayesian inference.
GAs usually use very heuristical goodness and offspring creation and selection methods. If they solve your problems that's obviously good enough for you but I think bayesian methods are better because:
When using bayesian inferece you know what you are doing (if you have studied lots of statistics). With GAs you don't and you are only guessing. GAs are good approximations of bayesian inference but one should concentrate on baysian inference instead. Any far reaching research on GAs probably only duplicate efforts usually already made by statisticians.
Say you want to find out probable good values for x (optimization, something else).
Define p(x) (starting population in GAs)
Define p(d|x) (related to goodness function of GAs)
sample from p(x|d)~ p(d|x)*p(x) (natural selection).
this is information theoretically optimal (H(X|DATA) = H(X) - I(X,DATA) so it isn't surprise that natural selection does something very close to this.
Why bother with GAs. They just special cases of baysian inference. Say you want to find out probable good values for x (optimization, something else). Define p(x) (starting population in GAs) Define p(d|x) (related to goodness function of GAs) sample from p(x|d)~ p(d|x)*p(x) (natural selection). this is information theoretically optimal (H(X|DATA) = H(X) - I(X,DATA) so it isn't surprise that natural selection does something very close to this.
Yes. It is also predicted that US dollar will drop -30% within year. Now ask yourself if it would be smart move to exchange all your savings to EUROs, wait 1 year and get 30% profit.
Chances are that even if you can remove the causes of ageing which can be notices when people are 200 years old, there may be other ageing effects which don't show up until you are more than 500 years old.
I'm not so sure about defacing. SCO has gone so nuts that it wouldn't surprise me if the pages were actually real new ones.