Other architectures

COMPETITIVE LEARNING

The output of one pRAM may be used as a penalty input to another pRAM.
In this way, competition between neurons or mutual inhibition can be achieved.
Similar mechanisms for the reward inputs can be constructed.

   

A neighbourhood of pRAMs can generate complex penalty inputs to adjoining pRAMs through the use on non-learning pRAMs which allow the penalty input to be a non-linear function of the neighbouring pRAM outputs.
A similar arrangement can be made for the reward inputs.
The non-learning pRAMs act as look-up tables.

More data

T G Clarkson, August 1995