In the laboratories of Western Sydney University, in the Australian capital, a group of researchers is putting the finishing touches on a new supercomputer. It’s called DeepSouth, and when it powers up next year, it will be the first machine capable of emulating the information processing capacity and speed of a human brain.
The machine has been designed with components that mimic the functioning of the synapse (the connection between neurons). It is capable of emulating about 228 billion of these connections per second, a capacity that those responsible for it believe is very close to what the brain has. It has been designed, in fact, for that sole purpose. “It operates exactly like a network of neurons,” explains André van Schaik, director of the International Center for Neuromorphic Systems, the university department in charge of its maintenance.
The equivalence is not direct. The human brain has about 100 billion neurons, capable of sending about 150 trillion signals. At first glance, DeepSouth’s capacity seems greater, but neurons can send more than one signal per second and there are different types of signals and connections that can be made. Still, DeepSouth’s power is high enough for experts to consider it equivalent.
DeepSouth, named after TruNorth, the first chip designed to imitate the behavior of a set of neurons and developed by IBM, will not, in any case, be as efficient as the Cerberus, which barely spends 20 watts of power to perform these tasks. connections, but it represents a leap in efficiency compared to supercomputers that have tried to do something similar using traditional computing components.
“Our understanding of how brains compute using neurons is hampered by our inability to simulate networks at this scale. Simulating neural networks on standard computers using graphics processing units and multicore CPUs is simply too slow and power intensive,” he explains. van Schaik.
Although DeepSouth will be the first to reach this scale, it is not the only project that has tried to replicate the connections of the human brain. The Human Brain Project, a European Union project announced in 2013, pursued a similar objective, but after more than one billion euros of investment and a decade of work, it was closed in September of this year without having achieved its objective.