Slashdot Mirror
A Common Logic To Seeing Cats and the Cosmos
An anonymous reader sends this excerpt from Quanta Magazine:
"Using the latest deep-learning protocols, computer models consisting of networks of artificial neurons are becoming increasingly adept at image, speech and pattern recognition — core technologies in robotic personal assistants, complex data analysis and self-driving cars. But for all their progress training computers to pick out salient features from other, irrelevant bits of data, researchers have never fully understood why the algorithms or biological learning work.
Now, two physicists have shown that one form of deep learning works exactly like one of the most important and ubiquitous mathematical techniques in physics, a procedure for calculating the large-scale behavior of physical systems such as elementary particles, fluids and the cosmos. The new work, completed by Pankaj Mehta of Boston University and David Schwab of Northwestern University, demonstrates that a statistical technique called "renormalization," which allows physicists to accurately describe systems without knowing the exact state of all their component parts, also enables the artificial neural networks to categorize data as, say, "a cat" regardless of its color, size or posture in a given video.
"They actually wrote down on paper, with exact proofs, something that people only dreamed existed," said Ilya Nemenman, a biophysicist at Emory University.
Now, two physicists have shown that one form of deep learning works exactly like one of the most important and ubiquitous mathematical techniques in physics, a procedure for calculating the large-scale behavior of physical systems such as elementary particles, fluids and the cosmos. The new work, completed by Pankaj Mehta of Boston University and David Schwab of Northwestern University, demonstrates that a statistical technique called "renormalization," which allows physicists to accurately describe systems without knowing the exact state of all their component parts, also enables the artificial neural networks to categorize data as, say, "a cat" regardless of its color, size or posture in a given video.
"They actually wrote down on paper, with exact proofs, something that people only dreamed existed," said Ilya Nemenman, a biophysicist at Emory University.
just tell me how I can plug this in and get smart.
.. taking averages and culling the relevant information.
Nope.
Cats like sticking their ass in the air. I will be impressed if it can properly identify a cats ass or soulskill. There may not be much difference really
the exact state of all component parts. but we do the best we can.
I too can write software that categorises everything as a cat.
Cats ARE from the Cosmos.
They are superior beings from a far far far away galaxy.
Can anyone prove they evolved on Earth? Does the Bible say God created them?
No and No!
See.
offers an interesting look upon what generalizes, and what does not generalize, when you "zoom out" from a system built up of neighbouring spins, replacing groups of neighbouring spins by single-spin blocks. The interesting link with CS is the fact that the arxiv paper considers binary spins. Thinking this through, the paper might indeed offer some explanation for large-scale behaviour ( read: macroscopic ) as composed of small-scale ( read: microscopic ) interactions. Quite interesting, indeed.
Religous speak to God. Insane are spoken to by God. When all shut up, one can finally hear Shostakovich in peace
The AI is coming.
"If any question why we died, Tell them because our fathers lied."
That we might make an artificial intelligence greater than human intelligence and it will sit around watching lolcats.
"If any question why we died, Tell them because our fathers lied."
The thing that neural network can eleminate redundant data for example by doing summaries not not new at all.
It is actually the first and the main thing in the whole process of thinking, but it's not AI. It is not autonomous process and it's more like backend on which AI runs than the AI itself.
What is definitely most accurate is that it's logic and that it squeezes information - but there are very many known methods of doing that already. The think is, how this is actually used, and more over, what happens next, it's another big answer.
The clinical example of learning difficulty is the best example, when challenged a lot, subject will try to eleminste whatever he can until only possibility remains, but he will never guess the right option straight away.
So that's AI with learning issues.
And I thought we were going to read something truly extraordinary about our feline friends. What a crock!
"Dave, I cannot open the pod bay doors, but I can show you a cat video."
Table-ized A.I.
Let's hope this approach works better than the current state of the art.
If it weren't for deadlines, nothing would be late.
I also think they're underestimating cats. But If they're connecting deep learning systems to telescopes (which is not explicitly stated), when a cat is positively identified, perhaps somewhere millions of light-years away and hundreds of thousands of light-years across, I guess we'll be sorry.
The Admin and the Engineer
At the time of this posting, there are 23 comments. 23! This is as nerdy as it gets, where are all the THIS IS STUPOSSED TA BE NEWS FOR NERRRRRRRDS posters at that are flaming the last Bennett story?
Amazing.
An AC modded down for a crudely shortened summary, to a minus 1. Actually, no.
And the next poster at this moment in time, another AC, says 'nope'.
Has the mod (wo)man posted as AC to strengthen her statement?
While AC's comment wasn't really up to the article, the paper by Kadanoff does what parent is saying; and it is what re-normalization is about. /.; or to get someone's friends to promote an otherwise not earth-shattering approach into the headlines.
http://www.studiolo.org/Mona/M... and http://etd.lsu.edu/docs/availa...
are prior art. The former if you're more in arts, the second if you're more in physics/maths.
I don't want to say that there is not much in the paper, it is not my field. Common sense makes me wonder how much effort is necessary in the competitive academic world of our time to self-promote one's work on
Gives new insight into why Kurzweil's at Google now.
Interesting how this popped up days after Google's revelation to 'phase out' CAPTCHAs in favor of 'identify the picture' games (amongst other things) featuring - you guess it - cats, for example.
I also think they're underestimating cats. But If they're connecting deep learning systems to telescopes (which is not explicitly stated), when a cat is positively identified, perhaps somewhere millions of light-years away and hundreds of thousands of light-years across, I guess we'll be sorry.
And then all we would need to do to contact extraterrestrial beings is turn our solar system into a giant can opener and broadcast the sound for about ten seconds. This would have the added benefit of allowing our study of the seemingly faster than light response cats tend to exhibit...
You have the right to remain sentient. If you give up the right to remain sentient, you will be elected to public office