Skip to main content.

How many useful states?

A high level view of the nervous system is given below. The interesting aspect is that mostly it talks to itself - sensory inputs and motor outputs are under 1% of the connections.


Consider a smaller network with 1000 neurons with all-all connectivity, about the size of a cortical microcircuits (although estimates of size of a distinct computational unit vary +/- an order of magnitude).


How many "useful" states is this network in? The early work of Hopfield assumed such a network would operate with a small number of "attractors".


For the purposes of this pattern machine, I interpret a 'state' as an instantaneous sparse activation pattern, giving 1000 C 10 possible 'states', also known as 'lots'. In practice, not all of the states will be distinguishable, and the limiting factor for capacity will be the number of useful and distinct transitions which can be stored.