Brain Research Corporation

Breinigsville, PA, July 07, 2009 --(PR.com)-- Brain Research Corporation is developing a hardware neocortex simulator capable of modeling a full human brain, consisting of 30 billion neurons. Completion of the simulator is set for two years from now and is expected to exhibit the full intellectual behavior of the human. Prediction of success of the simulator is predicated on the use of neuron nets as observed in the human cortex with the added (the unmeasured) constraint that the connections between excitatory and inhibitory neurons is large enough to guarantee that the inhibiting neuron will pulse if the connected excitatory neuron pulses, resulting in groups of neurons being mutually inhibiting. Adding a second (unmeasured) constraint that the inhibitory and excitatory connects vary by the same rule, results in the nets of mutually inhibiting neurons doing “normalized correlations” where the neuron that pulses is the one whose connection pattern best matches the input pattern. Adding a third constraint that the output of a neuron is its frequency of pulsing, not the pulsing waveform itself, results in a neuron net structure anticipated to exhibit human behavior at a fraction of the 30 billion number of neurons. These deductions are based on an electrical engineering principle that correlations are of little use unless they are normalized (the sum of squares of the excitatory connections is constant over all neurons, as the connections vary due to learning). The discovery that neurons are able to do normalized correlations (with only the connection values being constrained in order to cause normalization), makes it very likely that the cortex is using normalized correlations. Otherwise the nets of neurons would not perform useful pattern recognition behaviors. The modeling of a neuron with a frequency output, rather than a pulsing (at a frequency) output, has the advantage that many less neurons are needed for any useful behavior (it takes many pulsing neurons of a given frequency to produce a steady pattern input to a receiving neuron for pattern recognition purposes).

It is expected that the simulator will quickly evolve toward human behaviors as the trials are made to seek the major cortical to cortical connection patterns of the human brain, patterns which presently are only vaguely known or measured. The major effort of the project is to seek the cortical to cortical connection scheme and the initial values of the connections. Failure is not feasible, since the simulator can be reprogrammed to any connections, initial conditions, and neuron properties.

Using today parts and costs, the simulator will fit into a 2 foot cube and cost about 1 million dollars. This cost and volume will drop each year based on normal integrated circuit evolution of a factor of nearly 2 every year. Also, it is expected that 30 billion neurons is not need for useful intellectual behavior, so that the cost and size can be much lower for present practical applications.

One of the first uses of the simulator will be to serve as the brains for super robots for service in the military. The costs for these robotic soldiers will be significantly less than that of the human soldier, the numbers needed greatly reduced, and the performance greatly increased (these robots will be difficult to stop or kill.)

Although the intellectual performance of the neocortex simulator will be able to exceed that of the human, the units produced will be constrained to keep the machines in certain areas of endeavor so that the human is not displaced unnecessarily. The market size for the simulators will exceed 100 million units (world wide) in the near future for commercial purposes. Obviously the greater intellectual performance will be useful in research areas as well. Certain laws will be needed to keep these machines from replacing all humans.

Although funding for the project is from the two inventors of the simulator, in order to speed up the developments, funding is being sought from NSF, DARPA, corporations, and private investors.

Inquiry to this project can be made through email address, rswallow@ptd.net. Doctors Ronald J. Swallow and Richard L. Swallow are the principle investigators.

###