Inside the minds of the thinking computers

Mapping the brain

Before building its brain-based computer, IBM will have to understand, document, detail and map the workings of a brain. However, IBM is by no means the only company carrying out this research. In laboratories around the world, groups of computer experts and scientists are attempting to map the brain's workings in order to create this incredible new breed of thinking computers.

The most notable study – the Brain Atlas project – seeks to map the human brain, and it's occurring at the Allen Institute for Brain Science in Seattle. Paul Allen, the elder statesman of computing who partnered with Bill Gates at an early age to start Microsoft, has provided most of the funding. Allen was diagnosed with Hodgkin's disease in 1983, and he has worked as a philanthropist for the past several decades, with a decided interest in neuroscience.

The Brain Atlas project is amazingly ambitious, and not just because it seeks to create a complete data model of the genes at work in the human brain. The project also intends to make this brain atlas widely available and accessible over the Internet; to make it a kind of social network for scientists and researchers who, working together, can make scientific discoveries in ways that would not be possible if the Brain Atlas existed only in a closed lab.

The initial step was mapping the brain of a mouse. Using complex imaging techniques, the Allen Institute took over one million high-resolution scans for about 20,000 genes of the mouse brain, capturing the gene expression inside the brain. Next, they created ways for the research scientists to view and use the data. Currently, the institute has used 150 CPUs – each with 8GB of RAM – for the mouse brain project. The human brain is 2,000 times larger than the mouse brain, so if the Brain Atlas project scaled with hardware, it would require 10 terabytes of storage. The institute plans to use software and less memory per gene expression to make smaller scans so that they can be more easily distributed over the Internet.

"The bulk of our architecture is in storage, with 600 terabytes of data currently for the mouse project," says Chinh Dang, the Technology Director. "Each 2 x 3cm slide of a human brain, which is roughly 1/65,000 of the total brain, generates an image that ranges from 1GB to 2GB, and we will acquire and process millions of these images."

What's in a picture?

The other challenge with brain map imagery is that a single picture does not show the whole story. The brain has millions of overlapping fibres that exist within the same space, and every individual fibre has an important function inside the brain – for example, holding memory, providing pain response or solving a problem. In the past, brain scientists have surmised that the brain has specific sections, but the Massachusetts General Hospital research shows that the fibres overlap and pass information between sections.

"We are constructing images that show the fibre pathways of the brain," Wedeen explains. But things aren't quite as simple as they might sound. "It's an unusual data set because the pathways overlap three-dimensionally," he adds. "They are not even remotely visible all at once."

Wedeen says that the missing pieces of the puzzle – the reason brain mapping is still a new concept and not widely used outside of a research lab – has to do with graphics processing. The map is drawn by capturing the data from the brain, assembling the data and reading the fibre data. A program called TrackVis reads the MRI data and creates the brain maps. The construction can take a few minutes or several hours, depending on the complexity of the data and how it will be used.

In the future, GPUs will do even more: "Significant bottlenecks will be improved by moving to GPUs," says Wedeen. GPUs that would otherwise be sitting around unused will be harnessed toward the more mundane tasks of computing real data. "The other big issue is storing multi-gigabyte data sets in memory and accessing them randomly and rapidly. But medical imagery can parallelise easily on GPUs at a low cost."

Reading data from more than one brain is the next phase of the project – comparing two brains or even 10 simultaneously. One issue is that current displays are not capable of showing all of the detail of a 3D map of several brains – there are just not enough pixels. Today, 30in screens at a 2,400 x1,600 resolution are just not sufficient because they do not show all of the 3D relationships. So what's the Nirvana state? A 3 x 6K monitor would provide the best data visualisation for brain mapping.

Other benefits

The benefits of all this work could be manifold. If the IBM-led team were to fail in their lofty aspirations, the ripple effects created by the research is still likely to benefit all of us. Huge amounts of conventional computing power will be needed to understand and map the brain. This can only help our collective understanding of techniques such as parallelism and data visualisation.

And there's the more human benefits, too. The results of brain-mapping projects will transform the diagnosis and treatment of diseases such as MS and Alzheimer's disease, and could assist doctors to evaluate schizophrenics and those with serious drug addictions.

They could even help us to understand the differences between men and women – and if that isn't reason enough to cheer on cognitive computing research, we don't know what is.

-------------------------------------------------------------------------------------------------------

First published in PC Plus, Issue 278

Now read How to build a powerful distributed computer

Sign up for the free weekly TechRadar newsletter
Get tech news delivered straight to your inbox. Register for the free TechRadar newsletter and stay on top of the week's biggest stories and product releases. Sign up at http://www.techradar.com/register

Follow TechRadar on Twitter