A “LEARNING COMPUTER” that will help to solve the problem of how the human brain stores information is being built at University College, London,and should be in operation about a year from now.
The approximate electrical circuits of individual brain cells or neurons has been known for several years, and can be simulated on a conventional analogue computer using adders, multipliers, integrators and function generators.
One neuron requires at least 10 multipliers for its simulation, and it is clearly out of the question to simulate networks containing more than several dozen by this method.
When this is done the circuit is reduced to one adder,one multiplier and one integrator per neuron.
It is, however, only necessary to multiply by an integryal and these two functions have been combined in a single integrating-multiplier or I.M.unit.
The design of this unit has been a major part of the project and now that this has been achieved it will be possible to produce the several thousand required in the”learning computer”.
With this number available it will be possible to synthesise the elaborate computer set-ups re- quired for investigating new theories of learning nerve networks.
Source: electronicsweekly
