MIT brain bot mimics humans to recognize street scene
If imitation is the sincerest form of flattery, robots have been laying it on thick with human-inspired forms since day one. Unfortunately, AI hasn't quite kept up with mechanical achievements, mainly due to the fact that the human brain isn't merely mysterious in many regards, but dang complicated to boot. However, some brain experts at the Tomaso Poggio lab for brain research at MIT have made some impressive gains in bridging the gap. Using a computational model of how the brain processes info, the researchers created a bot capable of recognizing different objects in a street scene. The research is primarily targeted at brain geeks looking for ways to repair damaged brain functions and to better understand the brain's operation, but computer science should benefit from the biological inspiration of this research, and the methods used could lead to computer vision systems capable of better surveillance, car driving assistance, visual search engines and vision for robots. Because of its biological nature, the Poggio lab vision system is trained over time by being shown various street scenes, and the data in each picture is processed in much the same process as the brain goes through when presented with an image. Right now the system just mimics the brain's instant recognition faculties, but soon it'll be capable of putting more thought into what it sees -- for instance, cars do not belong in the sky -- so be warned people of earth: the first step in destroying an enemy is visually recognizing them. Our days are numbered.
[Via Medgadget]
[Via Medgadget]



















I like how the construction sign is mistaken for a pedestrian. We still control our own destinies...for now at least.
I for one... oh it's too easy.
what he said...
You see, Carl Marx was wrong about communism. Because there is always one thing that will be scarce in the human society regardless the level of productivity, and that's call "ideas".
But Carl Marx was only wrong of communism in a human society, in a broader context, ideas can be mass produced (as this post implies) and we will have communism eventually. The only thing here is, by the time communism is achieved, we (humans) we be eliminated by Robots/AI according to the evolution theory of Darwin.
These damn robots are getting smarter everyday. I wonder if it sees a pig in the sky if it would go haywire.
Good call Foof. Start stockpiling street signs now. It's the only way we will stand a chance.
"the brain's instant recognition faculties"
brains learn too. even recognizing an object as an object is an act of perception, not raw sensation, and involved employing some conceptual framework that recognizes distinct objects as such.
3rdsun
"I wonder if it sees a pig in the sky if it would go haywire."
Haha I think most people would. Metal or Meat brains alike.
It may be complicated to boot, but at least it doesnt crash very often {ducks}
Can someone please photoshop some crosshairs? It is just a matter of time before KILL-O-VISION V2.0 comes out.
Aleksei Efros at Carnegie Mellon is already doing this...http://www.ri.cmu.edu/pub_files/pub4/sivic_josef_2005_1/sivic_josef_2005_1.pdf
Eric,
The mass production of ideas is covered extensively in the book Accelrando by Charles Stross... and it really has nothing to do with Communism. Are you a time traveller from the 50's? Because other than you being from the near past, I can't imagine why anyone would care about "reds"
This defiantly is a turning point for the future computer based vision systems. I see military applications, partially surveillance applications, being the first to benefit from this. This coincides perfectly with the multi-core advances in chip design, which will enable CPU intensive “Biovision Systems” to compute high-detail live video. For instance, imaging reconnaissance video from a circling helicopter being analyzed instantaneous using this technology and identifying a particular target, say a man holding a rocket launcher, in real time. Of course, this is just one of numerous applications for a technology such as this.
Brian Glassman
www.TechRd.com