It sounds like a daunting task, but some researchers at
Harvard and
MIT have banded together to basically "reverse engineer" the human brain's ability to process visual data into usable information. However, instead of testing one processing model at a time, they're using a screening technique borrowed from molecular biology to test a range of thousands of models up against particular object recognition tasks. To get the computational juice to accomplish this feat, they've been relying heavily on GPUs, saying the off-the-shelf parallel computing setup they've got gives them hundred-fold speed improvements over conventional methods. So far they claim their results are besting "state-of-the-art computer vision systems" (which, if iPhoto's skills are any indication, wouldn't take much), and they hope to not only improve tasks such as
face recognition, object recognition and
gesture tracking, but also to apply their knowledge back into a better understanding of the brain's mysterious machinations. A delicious cycle! There's a video overview of their approach after the break.
[Thanks, David]
posted December 4th 2009 9:29AM
Open source knowledge is awesome!
@Jake88 Robotic ceiling cat is watching you masturbate for real now.
The brain works by building up sequences of linked neurons to form relationships and 'concepts'.
The neurons don't hold information in the same way a variable in a computer program does. It's the pattern between them that counts.
If they can work out exactly how this system works, they will find it can do a lot more than just recognise objects.
@(Unverified)
Actually, individual neurons to hold information. Their depolarization events set of protein transcriptions that alter the baseline state of the neuron, sensitizing or desensitizing the neuron to future events, or guiding internal event processes upon further depolarization. Neurons are not just wires.
The real trouble is that computers apply equal value to all fields of an image. The brain applies more value to items central to its field of focus, to items that are in motion (in specific directions, not just any motion), and to items with a stark contrast from their background. Shapes don't hold much value, except for the shapes of facial proportion (faces, cars, animals, classic houses). Applying these simple rules doesn't acquire a neural network, although my programming abilities are fairly non-existent these days. If a fourth dimension of "relevance" were added to each pixel as compared to its value in adjacent frames in conjunction with current facial proportion recognition, could produce a much more effective video processing application.
James Sonne, Ph.D. Student of Anatomy & Neurobiology, UK College of Medicine.
@James Sonne
Actually memories are formed from little angels who shape clouds in to pictures and sight is magic light which God splits, like rainbows, into picutures.
-James, home schooled with the teachings of the Church
@James Sonne
I didn't mean neurons don't hold 'any' information, just that it takes a sequence of them to represent anything meaningful. The biological processes involved are indeed complex, but much of this is unnecessary when you are making a computer model, as long as you understand 'why' it is happening.
@(Unverified)
(map
(lambda (stuff)
(... works by linking sequences of ... to form ...))
(
( programs , operations , informations )
( chips , gates , informations )))
\o/
Reminds me of "the borg"
Assimilation! ma hoy vin haven, with the lasers and the, voot voot
awesome stuff
1...
2...
3...
...Begin Robot Takeover
@greenskye
...Begin Robot Takeover of all beautiful things
I can see one problem with this approach: at least intuitively, our vision system is integrated on all levels - i.e. from the very highest 'A chair usually has four legs' to the lowest 'High intensity gradients are probably where edges of things are'.
I think encoding our high-level knowledge in the model will be difficult.
finally HAL comes alive
Anddddd were all dead
That way computers can discriminate based on looks too. :)
@One Love
Cigarette in mouth. Nicotine addict or poser.
@(Unverified) Ubuntu simble. Wannabe linux user or poser... wait, those are kinda the same huh?
@Brokinarrow aaand i fail at spelling : *symbol
@(Unverified)
That's not... exactly a "cigarette."
@(Unverified)
Ah yes, Marley, the greatest nicotine addict and poser of all time.
Learn some pop culture and how not to be a judgmental ass and then come back.
The best way to do this is to use memristors. A traditional transistor used in TTL logic is either on or off with a very brief in between period that isn't desired. Memristors change their resistance based upon the former current flow, so they can have any value from minimum to maximum and everything in between in their memory. This more perfectly replicates the functionality of the synapse which isn't just on or off but also gradations in between.
You aren't going to use binary systems for this kind of thing eventually, you need more levels than 0 and 1.
@uberFu Oh, you mean kinda like this: http://en.wikipedia.org/wiki/Memristor
And these: http://www.google.com/search?q=memristor&ie=utf-8&oe=utf-8&aq=t&rls=org.mozilla:en-US:official&client=firefox-a
Yea, seems like I'm the only one using the term. I feel so left out =(
@Wwhat Exactly, that's why I think memristors are the way to go.
@uberFu Oh my, apparently, even the man who originally theorized the memristors feasibility in 1971 even called it a memristor: http://www.youtube.com/watch?v=QFdDPzcZwbs
My world is becoming bigger!
Let's not do that.
great, there goes cyberdyne
OSI model?
I'm still waiting on the paradigm to shift on how we use computers. We use them now as tools, but i want my to act like a butler, or assistant.
Me: Hello Operator.
OP: Hello, Truth What Can i do for you.
Get me all the rss feed from all my favorite blogs and arrange them in order of comments and extract that boatload of German midget porn we downloaded yester...err
In all seriousness, when this does become mainstream, i would like to see general computing head this route (wont eve replace manual programming or creative task of course)
Correction: They use us as tools now.
Let's be honest for a chance :/
@(Unverified)
CS researches in intelligent agents are working on that. They're not necessarily perfecting it, but they are.
Maybe it's just one big cycle...humans invent AI robots, robots take over mankind, robots destroy mankind. Robots try and invent a more efficient, organic way of computing - invent humans as the solution - big battle - Skynet, Cyberdyne, Wargames etc...etc... :)
Always with the face recognition, can students stop making 1984 more doable and let the fascist think up their own crap already? Thanks in advance, when you are working on visual recognition recognizing a face is the last you should have on the list.
@Wwhat
Except the face recognition is on the top of the list of things that the human brain is capable of. Its power to recognize facial patterns and discern minute differences is unparalleled in any other computational device we know of. And so by copying it, we learn how it's done, so that we know more about our brains and about the world in general, and can take that knowledge and better apply to it other questions.
Researchers will keep on researching the world, and you can go on being paranoid and sending angry letters to your politicians. It's not scientists' job to make sure their tech isn't abused after they've learned about it, it's just their job to learn more about the universe. Einstein hated what his atomic research was used for, but he didn't stop, nor should he have.
Actually the whole facial recognition in humans involves a whole dedicated part of the brain and is much more specialized than you might think, and you need to learn to walk before you can run.
They know it's a whole separate parts and subsystem because there have been cases of people with physical brain damage who had all faculties and completely working vision but they became completely unable to recognize faces, and of course later more evidence came from modern scanners that show brain activity when doing tasks.
So no, to start with facial recognition is not the logical or scientific approach, and even if it was, sometimes you have to be wise and not give authorities tools that you know will be used only for nefarious uses, even if it means a slightly longer road for you to travel
It's like taking the time to turn the handle of a pan on the stove away from the reach of small children if you have kids, yeah it's easier to not do it but come on.
Schools should teach students a bit about responsibility too I think.
I get the feeling that whenever AI first comes online, It is going to visit Engadget to learn how to build the rest of itself.
It will subsequently hack into MIT, DARPA, NASA, etc. and download the construction plans it needs.
Artificial Robot hand.....check
Simulated Visual Cortex....check
Track-like tank feet.....check
Johnny 5....IS ALIVE!
Since it will have nvidia powered brains will it be hating on intel?
Stay tuned to find out.
P.S. Log out and back in to get the 'pick a name' box and rid yourself of 'unverified'
Why on Earth does a computer need to see? Skynet is coming people, Skynet is coming!!!
"You can't be sure you've build the plane correctly until it takes off and flies"
JESUS! That's why this guy is a neuroscientist and not an aerospace engineer!
Boring... Discover magazine covered this October.
http://discovermagazine.com/2009/oct/06-brain-like-chip-may-solve-computers-big-problem-energy
Actually similar research has been done for many many years, in many institutions all over the world, and some of it is in use already.
And that article doesn't seem the same at all on cursory glance, it's stanford not MIT & harvard for starters.
You can't shuffle all under one post 'scientist are investigating stuff' more news if they stop doing that..."
@lazuline while I have to agree this isn't terrible exciting, as off the shelf systems will only take them so far, the research aims between this group and the group you posted are quite different.
I was recently at IEEE: biomedical circuits and systems conference in beijing a week ago and there were some interesting talks on development of analogue computers. Basically simulating neurons in silicon. Very interesting stuff.
If anyone is interested do a search for Prof. Ralph Etienne-Cummings from Johns Hopkins University, or Dr. John V. Arthur from Stanford University.
"A delicious cycle!"
I think the Engadget editors are actually zombies.
No no no. All wrong. It's: Eat. Sleep. Sex.
They completely ignored the preprocessing done in the eye by ganglions, like detecting lines and shapes and contrast areas, their simplification of that all is sent as a simple picture is dooming the whole project, unless they just simplified that in the video for us dummies? Otherwise they should really get back to the drawing-board and get a neurologist/eye expert on the team.
Now if they would recognize that the human brain is not organic but energetic that would be a great advancement :D
@Marcel2097
Is you a girl?
human cannot create a machine that is like human.
Artificial intelligence is just a joke.
@xiaobai
Yes they can, just make an artificial intelligence, then spill a glass of tea on the circuit to dumb it way way down, and done.