Slashdot Mirror
Data Mining Used to Create New Materials
Roland Piquepaille writes "MIT researchers have successfully integrated data mining tools and modern methods of quantum mechanics. They've designed software which can help predict the crystal structures of materials. To simplify, they say they've used methods used by online sales sites to suggest books to customers. And it seems to work: they claim they can determine in days the properties of atomic structures that might have taken months before. Read more for additional references and pictures."
"its about time someone did something worthwhile with datamining ... i'm so sick of everything datamining is used for being big brother/1984 related."
That's just because the cynic grabs all the attention. Datamining has been used for years, by the fortune 500, and you don't hear much about that.
Oh J.S. Christ. For a forum of geeks you all know diddly squat about law. A software patent isn't the same as a business patent. Second you can't patent math.
"Using a technique called data mining, the MIT team preloaded the entire body of historical knowledge of crystal structures into a computer algorithm, or program, which they had designed to make correlations among the data based on the underlying rules of physics.
Harnessing this knowledge, the program then delivers a list of possible crystal structures for any mixture of elements whose structure is unknown. The team can then run that list of possibilities through a second algorithm that uses quantum mechanics to calculate precisely which structure is the most stable energetically -- a standard technique in the computer modeling of materials. "
No more freaking Roland Piquepalle!
This guy is way out there
"The second I'll call "intelligent observation"."
You know, most people would call it statistics (in this example, using a mathematical model to predict results), or the scientific method (in general, observing repeatable events).
--
Internet Explorer (n): Another bug -- that is, a feature that can't be turned off -- in Windows.
Wait, I thought that used to be Hemos?
And it's a sad thought that Zonk has really been around for years now...
Justice is the sheep getting arrested while an impartial judge declares the vote void.
Reading about how a program can find something that a human could not, or would not, brings to mind a notion I had the other day.
I am learning Java (and OO programming best practices in general), and am pretty heavily into it at this point. I was tooling along, writing some code to test some aspects of the language when I suddenly realized that much of what I was typing I was kind of unaware of.
When I had first begun studying in earnest a few months ago I remember how closely I paid attention to the smallest syntactical details. But now that much of this has become wrote I found myself automatically just cruising through - not really conceptualizing what I was doing. But it was still working.
I went back into my little code and delved into a deeper reading of what I had written. It was all correct according to theory - and I could recall all the little subtleties of how Java's VM was interpreting this and that - but while I was writing it I was giving no thought to it. It just happened; it just came out of me.
Now, hearing about these programs that can mine data and find things that human eyes would miss - and relatedly hearing about machines that can invent - I wonder if one day invention, discovery and the like will all be wrote.
I wonder if, like my mindless coding moment, things will just happen - research will just occur - without really a second thought of the "low-level" processes that currently are held so dear.
It's interesting. It might be akin to mathematics in some ways - wherein you can generalize a large body of calculation and come to a conclusion without actually outputting the raw numerical form.
It is an approximation, yes. But with some work the approximation can be decomposed into elementary school level math expressions - if you really want to go through all that work.
But why decompose it, it works fine generalized (much better for humans in fact).
It's interesting to me - this modern high-level generalization.
My Computer Music Tutorial Videos
... not.
But I did work on a project that applied data mining techniques to drug screening problems. Specifically, we used kernels on molecule data in a support vector machine to predict the outcome of AIDS and cancer screening data. It worked moderately well. (AUC of up to .94)
So: Surprise, surprise, data mining is used for all kinds of things! Drug screening, materials engineering, process control, analyzing NMR spectra, ... it's not just marketing! Basically, every application that produces a lot of data will eventually have data mining people flock to it, trying to data mine the heck out of it.
Regards, Sebastian
No chemist will ever trust a computer result without doing the full lab work. This can still be incredibly useful.
Consider if each thorough test takes 6 months for 3000-4000 possibilities. If the computer can tell you the 5-10 compounds that are likely to work, in a few years you can have a product (or a PhD). Otherwise you were looking at nearly a thousand years before finding something.
... who has never actually used genetic programming. Genetic programming doesn't create new inventions -- it typically tweaks parameters in an existing invention so that the output of the invention approaches a goal. For example, you could use it dynamically weigh, say, SpamAssassin test scores. It doesn't just magically evolve new tests, and it certainly doesn't evolve a regular-expression based server side spam filter, it just tweaks the efficiency of one which already exists. Even for artifically restricted problem domains, such as CoreWars or similar combative programming environments, the successful A-life programs generally revolve around optimizing a strategy and a base implementation which a human came up with. Call it "intelligent design", because thats what it is :)
They also most certainly do not beat all existing algorithms. In some problem domains they work very well. In others (hmm, lets see: sort, calendar applications, Internet telephony, uncountably many fields of human endeavor) they're wholly 100% inapplicable.
Help poke pirates in the eyepatch, arr.