A study published in the open access journal Frontiers in Neuroscience found that a computer based on simulating neural networks in the brain performed similar to supercomputers running the best brain-emulation software used in neural signal research. When tested for accuracy, speed and power consumption, this unique computer, the SpiNNaker, has the potential to outperform conventional supercomputers in terms of speed and power efficiency. The aim is to expand knowledge of the workings of neurons in the brain, as applied to learning and disorders such as epilepsy and Alzheimer's disease.
SpiNNaker is capable of providing detailed biological models of the cortex (the outer layer of the brain that receives and processes information from the senses), producing results very similar to those obtained when running emulation programs on a supercomputer,”says Dr. Sacha van Albada, lead author Research and Team Leader for Theoretical Neuroanatomy at the Julich Research Center in Germany. "The ability to execute large-scale, detailed neural networks quickly and with low energy expenditure will contribute to robotics research as well as the study of brain disorders."
The human brain is very complex and contains one hundred billion interconnected cells. We have an understanding of how individual neurons and their components operate, and how they interact with each other, which areas of the brain are used for sensory perception, action and cognition. But we know less about the transformation of neural activity into behavior, such as how thought is transformed into muscle movement.
Supercomputer software has helped emulate signaling between neurons, but even the best programs on the fastest computers today can only emulate 1 percent of the human brain.
“It is not yet clear which computer architecture is best suited for efficiently running a whole brain emulator. The European Human Brain Project and the Julich Research Center have conducted extensive research to determine the best strategy for this daunting task. Today's supercomputers take minutes to emulate one second of real action, so research such as learning processes is not available today, explains Professor Markus Dizman, co-author and head of the Department of Computational Neurosciences at the Julich Research Center. - There is a huge gap between the energy consumption of the brain and the supercomputer. Neuromorphic (brain-like) computing allows us to understand how close we can get to the energy efficiency of the brain using electronics."
Developed over fifteen years and based on the structure and ways of the human brain, SpiNNaker - part of the European Brain Research Project's neuromorphic computing platform - consists of half a million simple computing elements. The researchers compared SpiNNaker's accuracy, speed, and energy efficiency to NEST, a specialized supercomputer software used to study neural signals in the brain.
"The emulations running on SpiNNaker and NEST show very similar results," says co-author Steve Furber, professor of computer engineering at the University of Manchester. - For the first time, such a detailed emulation of the cerebral cortex has been produced by means of SpiNNaker (or any other neuromorphic platform). SpiNNaker includes 600 boards combining more than 500,000 small processors. The emulation performed in this study used only six boards, which is 1% of the machine's full power. Our results will help improve the software and reduce the number of boards used to a single one."
As Dr. van Albada says, “We are looking forward to doing more real-time emulations using such neuromorphic computing systems. In the European Brain Research Project, we are already working with neuro-robotics specialists who hope to apply our findings to control robots.”
Promotional video:
Vadim Tarabarko
