Modeling a real brain

i-bc6905b9beff9f453ebefa20d6598e3d-compbrain.gifA rat's brain has millions of neurons, each with up to 10,000 connections to other neurons. This "simple" animal's neural network is mind-bogglingly complex. Yet a Swiss laboratory has achieved remarkable success duplicating a vast region of a rat's brain using a supercomputer. They still have a ways to go, however. The computer currently has 10,000 microprocessors, each representing a single neuron in the rat's brain. To duplicate the entire brain they'll need a computer 2,000 times bigger. Their ultimate goal is even more ambitious: to create a model of the human brain, with its hundred billion neurons.

Creating a computational model of a portion of a brain has been done before; what's different about this project is its insistence on absolute biological accuracy:

Researchers opened thousands of rat skulls over the years, removed their brains, and cut them into thin slices, which they kept alive. Then they directed tiny sensors at the individual neurons. They listened to the cells firing neurons, and intercepted the responses coming from the adjacent cells.

In the end, the researchers at Markram's lab collected the entire repertoire of behavior of hundreds of types of cells in every conceivable situation in a rat's life -- stored in endless tables.

So instead of starting from an observed behavior and attempting to come up with a computer model to replicate it, Markram's team observed the activity in the brain itself, and created its model based on that. Only afterwards do the look at their model and see if its activity matches real-world behavior.

The so-called neural networks that other researchers have developed for years are very different. A neural network is also meant to behave like a brain, somehow, but how the likeness is achieved -- the circuitry underneath -- is more or less irrelevant. Someone building a cow using this principle would be content with any milk machine that moos and produces the occasional cowpat.

"That doesn't help us to understand the biology," says Markram. The researchers in Lausanne are interested in the real cow: "Our first priority is that we never use tricks to achieve the correct result," says project manager Schürmann. "If something goes wrong in the simulation, our only option to improve it is by incorporating new biological knowledge."

Those are certainly high ideals. In future years, we'll see if the project can continue to live up to them.

(via Mind Hacks)

Tags
Categories

More like this

Last week, a team of computer scientists led by Dharmendra S. Modha announced what sounded like an impressive breakthrough for neuroscience-inspired computing: Using Dawn Blue Gene / P supercomputer at Lawrence Livermore National Lab with 147,456 processors and 144 TB of main memory, we achieved a…
Henry Markram, the director of the Blue Brain project, recently delivered a talk at TED that's gotten lots of press coverage. (It was the lead story on the BBC for a few hours...) Not surprisingly, all the coverage focused on the same stunningly ambitious claim, which is Markram's assertion that an…
The image on the right is a supercomputer simulation of the microcircuitry found within a column from the neocortex of the rat brain. The simulation is a tour de force of computational neuroscience: a single column is a highly complex structure, containing approximately 10,000 neurons and 30…
Via Vaughan at MindHacks, comes this link to a preview of a documentary-in-progress on The Blue Brain, that epic attempt to create a conscious supercomputer. I was fortunate enough to profile the Blue Brain in 2008: In the basement of a university in Lausanne, Switzerland sit four black boxes,…

"It had hardly been fed electrical impulses before strange patterns began to appear on the screen with the lightning-like flashes produced by cells that scientists recognize from actual thought processes."

I know it's English from the Spiegel article, but somehow it's not my language. How strange, both in degree and character? Lightning-like flashes? You mean action potentials? Gee, my own experience with those was in a context that had nothing to do with actual thought processes. Hopefully the scientists involved will give better descriptions of their results someday.

I agree; the translation leaves a lot to be desired. I think the translator didn't really understand the science involved. But reading between the lines, it's still interesting stuff.

Its amusing that this translation includes this, "...cut them into thin slices, which they kept alive. Then they directed tiny sensors at the individual neurons.... In the end, the researchers at Markram's lab collected the entire repertoire of behavior of hundreds of types of cells in every conceivable situation in a rat's life..."

How is having your brain sliced and poked with electrodes "every conceivable situation"??

By doublehelix (not verified) on 26 Mar 2007 #permalink

Doublehelix:

Heh. Hopefully that's another translation error.

Brian:

Thanks for the link. Yes, the project was announced in 2005. I thought readers might be interested to know the progress they are making.

Is there a reason they need to simulate this with hardware? That sounds expensive and prone to error (or more likely just more expense in order to ensure there are no errors.) I'm sure one could use software to simulate the function of computer hardware. I'm also sure they've thought this out a lot better than I have, so I'm curious as to the reasoning. :)

Roy, hardware simulators (and constrained, limited power simulators in particular) provide some insights you can't easily get otherwise. Working bottom-up under physical and other limitations change the way one thinks.

If all we wanted a working brain, a real mouse would work better (and be much cheaper). It's the craft of brain building that matters; we have too little experience in that department.

By On Paradise (not verified) on 28 Mar 2007 #permalink

Specialized hardware can be a lot faster and lower in power-use than using general-purpose chips for the same task. Like 100x faster, I think.