IBM says they have moved a step closer towards creating an artificial brain. I've been following their work, expecting their system to achieve consciousness. I may have been a bit hasty in my expectations. Of course, this is a tremendous leap in a gargantuan mission, but their system has artificial neurons (computation), synapses (memory), axons and dendrites (communication). The system dubbed, TrueNorth has been tested using two supercomputers: the IBM Blue Gene/Q (using 16 racks, 262,144 CPUs and 256 TB) as well as on Lawrence Livermore National Lab’s Blue Gene/Q Sequoia (using 96 racks, 1,572,864 CPUs, 1.5 PB memory).
TrueNorth is actually a cognitive computer architecture designed to simulate the creative and imaginative right-brain, in a computer. The goal is to perform more condensed, short and parallel computing with traditional computational, or left-brain, computers. The TrueNorth architecture is non-von Neuman, ultra-low power, modular and compact to simulate the scalable network of neurosynaptic cores.
IMB’s neurosynaptic cores, designed with help from Cornell, measure 3mmx2mm and contain 256 neurons and 1024 axons. These use SRAM crossbar memory and, according to lead researcher Dharmendra S. Modha, they are the key building blocks to construct “modular neuromorphic architectures."
The network simulated on the Sequoia used 256 million neurosynaptic cores that totaled 65 billion neurons and 16 trillion synapses. This artificial brain ran 388 times slower than the natural one and had an 8.1 Hz average spiking frequency. In order for TrueNorth to function, a functional simulator and compiler called Compass was developed. This multi-thread parallel program maps and simulates the long-distance pathways of a Macaque monkey brain.
Another test had IBM working with 2.084 billion neurosynaptic cores, totaling 530 billion neurons, 137 trillion synapses, but ran 1,542 times slower than what we all have in our heads. Despite the slowness, IBM’s brain can do some fascinating tasks like predict half written symbols and play games like virtual pong while its “sensory and motor” neurons are being monitored (though there is no way to tell whether its cheating or not). I will be impressed when it consistently beats people at real pong or chess, paints a mural or composes a song.
This project is part of DARPA’s SyNAPSE (Systems of Neuromorphic Adaptive Plastic Scalable Electronics) program which aims to build a simulator with 10 billion neurons, 100 trillion synapses (estimated number of synapses in human brain) to scale these systems to that of a human brain. DARPA has just funded the next stage in the project with $21 million dollars.
Here is what the IBM group had to say about their lofty project:
“We have not built a biologically realistic simulation of the complete human brain… Computation, memory, and communication are mathematically abstracted away from biological detail toward engineering goals of maximizing function (utility, applications) and minimizing cost (power, area, delay) and design complexity of hardware implementation. Previously, we have demonstrated a neurosynaptic core and some of its applications… We have also compiled the largest long-distance wiring diagram of the monkey brain. Now, imagine a network with over 2 billion of these neurosynaptic cores that are divided into 77 brain-inspired regions with probabilistic intra-region (“gray matter”) connectivity and monkey-brain-inspired inter-region (“white matter”) connectivity. This fulfills a core vision of the DARPA SyNAPSE project to bring together nanotechnology, neuroscience, and supercomputing to lay the foundation of a novel cognitive computing architecture that complements today’s von Neumann machines.”
The IBM group’s results were discussed at the International Conference for High Performance Computing, Network, Storage and Analysis in Salt Lake City, Utah in November. The SyNAPSE project is being worked on by IBM, Columbia University, Cornell University, University of California, Merced and the University of Wisconsin- Madison.