© 2008 vWISP PTY LTD

NeoCortex Processors

 

During the fall of 2003 vWISP set out to create a new processing architecture based on the human brain. Now, more than five years later, the NeoCortex-2 hierarchical neural processor is set to break into what will eventually be a Trillion-dollar market.

 

vWISP has created a dynamic artificial neuron (DAN) consisting out of ~4000 binary gates. The neuron exhibits STDP and BCM learning, temporal and spatial integration, neurotransmitter simulation, tonic and burst firing just like a biological neuron. Our brain does not contain any processors or memory devices. Everything the brain does has to be explainable from neuron physiology. Our next target is to put 15,000 cores on a single chip, which can be done using today’s technology.

A 15,000 core matrix executes a neural algorithm 60x faster than a Pentium Duo.

 

Each synapse is an event input. We have diverged from the ‘neural code’ idea. We treat every input as a temporal-spatial event. The neuron gathers knowledge from the environment that it is exposed to, by pulse timing plasticity. (STDP). This knowledge is stored as a set of dynamic values in the synapses.

 

Events trigger knowledge and update the synapse strengths through STDP. The output of each neuron is a probability event, resulting not from ‘logic’ but from temporal combination of knowledge. At the lower levels of the hierarchy, knowledge stored in synapses is simple; a visual line segment, audio frequency or something like that. Higher in the hierarchy the knowledge gets increasingly complex and comprehensive. At the highest level the output of different hierarchies is associated to determine meaning.

 

Applications for this type of machines include medical prosthesis, robotics, exploration systems, security systems that can recognize events to eliminate false alarms, interactive education systems, voice recognition, intelligent games, object and face recognition, fingerprint analysis and intelligent cars.

 

Details