PORTLAND, Ore.—Despite calls from some corners for more complex programs, the European Union’s Blue Brain Project said Monday (Sept. 17) that simpler supercomputer simulations are sufficient to realize its goal of creating an electronic brain.
Many neuroscientists had insisted that accurate brain simulations required mimicking the chemical-messaging that guides the growth of connections between neurons in the brain. But scientists at the Brain-Mind Institute at Ecole Polytechnique Fédérale de Lausanne (EPFL, Lausanne, Switzerland) counter that randomly wired neural networks derived from reconstructed neurons are up to 95 percent.
The Blue Brain Project’s goal is to simulate the neural networks of the brain on supercomputers, then cast these virtual circuits into semiconductor hardware once their accuracy is high enough. If the chemical messaging systems of what neuroscientists call the brain’s connectome— a comprehensive map of neural connections in the brain—had to be simulated, too, the task would have been incredibly complex, adding years to the project’s timeline. Luckily, according to EPFL, a randomly wired connectome is sufficient.
“We simulated the connectome by detecting the touches between randomly placed 3-D neurons and compared them to experimental results in real brains and found that that our structural model predicted the distribution of connections with sufficient accuracy,” said principle author of the EPFL report, Sean Hill.
IBM Blue Gene supercomputer simulates 10,000 of virtual neurons packed into a 3-D space in random positions according to the density and ratio of morphological types found in corresponding living tissue.
EPFL simulated on an IBM Blue Gene supercomputer over 10,000 virtual neurons packed into a 3-D space in random positions according to the density and ratio of morphological types of neurons in human brains. The team then compared its results to data from 20 years of experimental results about the geometric and electrical properties of living brain tissue. The results showed a 75 to 95 percent correspondence, with the exceptions indicating special cases used to change the network's statistical connectivity to make its operation more robust and immune to loss of function in any one neuron.
If EPFL’s controversial findings can be independently verified in coming months, then the ability to use random wiring topologies will also aid similar efforts at IBM, Hewlett Packard, HRL Labs and scores of European and Asian labs worldwide which are also constructing supercomputer simulations of neural networks in advance of constructing electronic brains.
The report, entitled "Statistical connectivity provides a sufficient foundation for specific functional connectivity in neocortical neural microcircuits," will be published today in the prestigious Proceedings of the National Academy of Sciences (PNAS). The video below explains the details.
Join our online Radio Show on Friday 11th July starting at 2:00pm Eastern, when EETimes editor of all things fun and interesting, Max Maxfield, and embedded systems expert, Jack Ganssle, will debate as to just what is, and is not, and embedded system.