Tuesday, 10 February 2009

Numbers and the Brain

The computational resources required to model the brain are huge - but within our reach soon. From Wired:

The structure of the cerebral cortex is the same in all mammals. So researchers started with a real-time simulation of a small brain, about the size of a rat's, in which they put together simulated neurons connected through a digital network. It took 8 terabytes of memory on a 32,768-processor BlueGene/L supercomputer to make it happen.

The simulation doesn't replicate the rat brain itself, but rather imitates just the cortex.
The human cortex has about 22 billion neurons and 220 trillion synapses, making it roughly 400 times larger than the rat scale model. A supercomputer capable of running a software simulation of the human brain doesn't exist yet. Researchers would require at least a machine with a computational capacity of 36.8 petaflops and a memory capacity of 3.2 petabytes -- a scale that supercomputer technology isn't expected to hit for at least three years.
Modelling the brain through sheer compuational grunt allied with software is just one approach that's being examined.
Software simulation of the human brain is just one half the solution. The other is to create a new chip design that will mimic the neuron and synaptic structure of the brain.

That's where Kwabena Boahen, associate professor of bioengineering at Stanford University, hopes to help. Boahen, along with other Stanford professors, has been working on implementing neural architectures in silicon.

One of the main challenges to building this system in hardware, explains Boahen, is that each neuron connects to others through 8,000 synapses. It takes about 20 transistors to implement a synapse, so building the silicon equivalent of 220 trillion synapses is a tall order, indeed.

"You end up with a technology where the cost is very unfavorable," says Boahen. "That's why we have to use nanotech to implement synapses in a way that will make them much smaller and more cost-effective."

Boahen and his team are trying to create a device smaller than a single transistor that can do the job of 20 transistors. "We are essentially inventing a new device," he says.
Which, if it comes to fruition, will have rather more immediate uses than modelling the brain.

The trouble is, that even if you do all of this successfully, all you manage to do is to model a mammal in a persistant vegetative state. Intelligence is not so much a thing as a process, and forming a mind given the components may just be as difficult a task as painting the Mona Lisa after inventing paint, canvas, and brushes. A whole new order of difficulty.
Meanwhile, at the University of California-Merced, Kello and his team are creating a virtual environment that could train the simulated brain to experience and learn. They are using the Unreal Tournament videogame engine to help train the system. When it's ready, it will be used to teach the neural networks how to make decisions and learn along the way.
Will the model of the brain be self-bootsrapping, self-ordering, the way that happens in nature? That's likely. But will it be sane at the end of it? That's another issue.
Then there's the difficulty of explaining that mimicking the cerebral cortex isn't exactly the same as recreating the brain. The cerebral cortex is associated with functions such as thought, computation and action, while other parts of the brain handle emotions, co-ordination and vital functions. These researchers haven't even begun to address simulating those parts yet.
It could be that these would be even more difficult to deal with, yet essential for any model worth the name to implement. This kind of tool though will be very useful in figuring out how the brain doesn't work, even ifthe ultimate aim is not achieved initially.

It's certainly our best, nay, our only hope of figuring out how we think, and how we feel. Just modelling the lymbic nucleus, and comparing the results with "normal" and near-normal but "anomalous" neuro-anatomy (for example, that involved with transsexuality) should tell us many of the things we want to know. Eventually. It might take decades, or even centuries though.

More on neural modelling in previous posts.


Anonymous said...

Neat stuff, Zoe. Query: Would I be able to understand your dissertation once it is completed, or would it be too technical for me to understand?

Anonymous said...

After the hardware you need the software.

Anonymous said...

"Then there's the difficulty of explaining that mimicking the cerebral cortex isn't exactly the same as recreating the brain."
And we're only in the very early stages of understanding the whole brain, see for instance: http://lanl.arxiv.org/ftp/arxiv/papers/0901/0901.4598.pdf
There's a nice reference on page 3 to the Allen Brain Atlas, reminiscent of the Alan Brain Atlas don't you think? Full of maps of a bygone era, I imagine. Pocket Scientist

VĂ©ro B said...

Funny, before I reached the last paragraph, I was thinking that even a model of the cerebral cortex needs a lymbic system model too. I should have known you wouldn't forget that. :)

Fascinating stuff!

mythusmage said...

How do neurons process information? Learn that and then you can talk about modeling a neuron. And when you can model neurons, then you can talk about modeling brains.

We so want AI we grossly underestimate the difficulty, and that will continue to stymie our efforts until we accept the fact we don't even know enough to model the information processing needed to build the simplest virus.